Chapter 13 retirement decision; current influences on the timing of retirement among older workers

288 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling13.4.6 Implication of Business Cycle for DC Plan This ch a pter i nvesti gat es t he i nfl uences on retirement behav

Gaobo Pang, Mark J Warshawsky,

and Ben Weitzer

CONTENTS

13.4 Empirical Results: Estimating and Explaining

* Opinions expressed here are the authors’ own and not necessarily those of their affi liations

We thank Wendi Bukowitz, Carl Hess, Richard Jackson, Allen Jacobson, Erika Kummernuss, Michael Orszag, James Poterba, Mark Ruloff , Ken Steiner, and seminar participants at t he Center for Strategic and International Studies for useful comments.

288 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling

13.4.6 Implication of Business Cycle for DC Plan

This ch a pter i nvesti gat es t he i nfl uences on retirement behavior

among older workers who were surveyed by the Health and Retirement Study (1992–2004) It is found t hat i ncreases i n a ll categories of wealth (pension, housing equity, and other fi nancial wealth) raise the probability

of retiring, while good earning prospects induce continued employment Retirement plan types have signifi cant impacts: workers covered by defi ned benefi t (DB) plans are more likely to retire, while the defi ned contribution (DC) plan coverage delays retirement Th e probability and thus the timing

of retirement for DC plan participants are susceptible to the infl uence of business cycles through retirement plan income fl ow fl uctuations that are due to investment performance and interest rate changes Health insur-ance (HI), if conditional on employment, strongly defers retirement, while alternative sources of insurance, such a s employer-sponsored retiree HI, spouse’s HI, public HI, or COBRA coverage, encourage labor force exit

Th e increases in the full retirement age for social security act to encourage younger cohorts to work longer

Keywords: Retirement, pens ion i nvestment, defi ned benefi t,

defi ned c ontribution, s ocial s ecurity, he alth i nsurance, bu siness cycle

JEL Classifi cations: J26, H55, J32, E32, H51.

13.1 INTRODUCTION

Th e fac tors a ff ecting work ers’ re tirement b ehavior h ave at tracted much attention among academia and policy makers Th is issue deserves renewed research attention and deeper understanding, given recent developments such as the decline in defi ned benefi t (DB) pension plans and the shift to defi ned contribution (DC) plans, ongoing social security (SS) reforms, and exploding health care costs for retirees as well as for workers

Retirement Decision ◾ 289

Th is chapter investigates the determinants of retirement behavior among older w orkers t hat w ere su rveyed b y t he H ealth a nd Re tirement S tudy (HRS, 1992–2004) Our analysis includes both conventional explanatory variables and new variables to refl ect recent environmental changes We revisit issues deemed important in previous studies and add new insights

to t he r etirement l iterature F irst, o ur d ata f ollows t he em ployment–retirement behavior of older workers for up to a dozen years Second, our modeling of the ongoing SS retirement age changes reveals the signifi cant policy-driven retirement diff erences ac ross co horts Th ird, we c ompre-hensively model all major sources of health insurance (HI) coverage and identify their varying impacts Th is approach may avert the omitted vari-able bias that could otherwise occur from examining individual factors in isolation Fourth, besides the fi nding of a signifi cant diff erence in retire-ment timing between DB and DC plan participants, our construction of the DC wealth-earnings replacement rate provides a unique way to gauge the susceptibility of DC plan participants to stock market and interest rate

fl uctuations

It is found that increases in all categories of wealth accumulation (retirement plan, housing equity, and other fi nancial wealth) increase the probability of retiring, but d iff erentially, wh ile good earning prospects, implying high opportunity cost for retirement, induce continued employ-ment It i s worth noting t hat our construction of e arning prospects or opportunity cost s forms a n a lternative but a m ore st raightforward a nd signifi cant way to incorporate the forward-looking incentives for contin-ued employment (such as the DB or SS benefi t accrual), which are shown

by some studies to be important

Retirement plan types have signifi cant impacts on retirement: besides the nearly universal SS, workers who are entitled to DB plan benefi ts are more likely to retire than those who are not, while the DC p lan cover-age delays retirement Th is phenomenon is presumably in part because many DB plans have work disincentives beyond certain ages while DC plans are largely age neutral, and in part because most DB plans provide

a more secure retirement income fl ow, lowering overall household sure to risk

expo-Th ere is a concern that the retirement behavior of the DC plan–covered workers is sensitive to stock market boom and bust Our analysis incorpo-rates business cycle eff ects, as part of the total DC plan eff ect, by includ-ing in come fl ow fl uctuations t hat a re d ue t o i nvestment per formance and ma rket i nterest r ate c ycles Th ese a re r isks pa rticular t o DC p lan

290 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling

participants We fi nd new evidence to support the above hypothesis; that

is, the probability and thus the timing of retirement for DC plan pants are susceptible to the infl uence of business cycles Workers who have signifi cant income loss (gain) in their DC plans are less (more) likely to retire Th is may impose some challenges to employers in workforce man-agement When there are market booms, DC plan participants retire just when companies need to add w orkers and when there are market busts,

partici-DC plan participants stay at work just when companies want to cut the workforce

Regarding the impact of HI coverage on the retirement decision, our study reveals that HI, if conditional on employment, strongly discourages retirement, while alternative sources of HI, such a s employer-sponsored retiree HI (RHI), spouse’s HI, or public HI, facilitate or may encourage labor force exit Th is fi nding highlights the importance for employers, in pursuit of strategies for human resource management, to consider retire-ment incentives inherent in pension plans jointly with the benefi ts pro-vided by other programs such as HI It should also be noted that benefi t modifi cations for retirees (such a s en hancing/eliminating re tiree he alth care coverage) may alter the retirement incentives for current employees.Various studies have investigated the importance of SS benefi ts as an explanation for early retirement In this respect, we have rigorously incor-porated the cohort-specifi c actuarial adjustment factors of SS benefi ts as defi ned by the law Our analysis fi nds that the retirement behavior is sig-nifi cantly linked to such public policies Th e ongoing increase in the nor-mal retirement age for SS will encourage younger cohorts to work longer

We c arefully i ncorporate va rious demographic cha racteristics i n t he regressions to control for the heterogeneity of retirement behavior, to the extent a llowed by t he d ata We however ack nowledge t hat our reduced form model may bear some insuffi ciency in addressing the probably simul-taneous determinations of savings and labor supply, joint retirement deci-sions of couples, job mobility, and the availability of pension and health care coverage, and other endogeneities Structural models have been used

by researchers to deal with the endogeneity issue and have advanced the understanding of re tirement b ehavior i n s ome d irections Th ese mod-els, however, a re oft en confi ned to one particular aspect of behavior or the environment due to their complexity, and furthermore bear the risk

of biased parameter specifi cations Reduced form models, which can be more transparent and comprehensive, ser ve as useful complements and guides to structural models

Retirement Decision ◾ 291

Th is chapter is organized as follows Section 13.2 reviews the relevant literature, Section 1 3.3 de scribes t he d ata, Section 1 3.4 d iscusses t he regression results, and Section 13.5 concludes

13.2 LITERATURE REVIEW

Th ere is a la rge literature exploring the potential determinants of ment a mong o lder w orkers Ma ny i nsightful t heoretical a nd em pirical research fi ndings ha ve co ntributed t o a de eper u nderstanding o f t his complex issue Yet, many debates and questions can be answered by new methods and data As a recent example, essays in Madrian et al (2007), all using the HRS surveys, examine retirement prospects, health status and

retire-HI, as well as wealth and asset investments for baby boomers Th is section highlights briefl y t he most policy-relevant strands of research and does not intend to make the review complete

One line of research has investigated the importance of wealth accrual and pension coverage on the timing of retirement Stock and Wise (1990) argue that workers have an incentive to remain in continued employment until certain ages (oft en the early retirement ages in pension plans) if the expected gain in utility from postponing retirement outweighs the value of immediate retirement Coile and Gruber (2000) examine the SS incentives for retire-ment and argue that it is in workers’ best interest to stay on the job so as

to maximize the SS wealth accrual—the “peak value.” Samwick (1998) also

fi nds that the accrual rate of retirement wealth is a signifi cant determinant

of the probability of retirement He argues that the rapidly growing pension coverage and SS entitlements since the 1940s could be the underlying cause

of the decline in labor force participation in the early postwar period

A second direction of research looks at the impact of pension types on retirement behavior Friedberg and Webb (2005) argue that DB plans tend to have age-related work (dis)incentives that fi rst discourage and later encour-age retirement, which contribute to early retirement and lead DB-covered workers to retire a lmost 2 y ears earlier on average, compared to workers with DC p lans Munnell et al (2004) study how pensions aff ect expected and actual retirement ages Regarding the actual retirement decision, they

fi nd that pension wealth increases the probability of retiring, while nities for more pension accruals lower the probability, and that DB coverage

opportu-per se raises the probability of actual retirement, while DC coverage reduces

the probability Based on survey data about faculty retirement expectations, Flaherty (2006) fi nds that individuals in DC-only plan situations expect to retire nearly a year later than those in a DB plan, in the context that earlier

292 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling

voluntary en rollments i n D B o r DC p lans b y fac ulty m embers t o so me degree reveal their diff erential retirement preferences

DB plans provide a steady stream of guaranteed income, while DC plans place participants in considerable exposure to investment and longevity risks Th e decline in DB plans and the shift to DC plans in the past decades have aroused concern that DC plan–covered workers are vulnerable to busi-ness cycles or stock market booms and busts Th e empirical fi ndings thus far are inconclusive, however Cheng and French (2000) estimate that about 15% of individuals aged 55 and over had an unanticipated wealth increase

of $50,000 or more in constant 1999 dollars between year-ends of 1994 and

1999 Th e labor force participation rates among them, however, increased

in t his ma rket boom period Th e authors bel ieve t hat t he r un-up i n t he stock market was not the primary determinant of employment changes in those years Th ey conjecture that the changes may well be a ttributable to the improved employment opportunities and wages in the strong economy and the reduction of work disincentives in the SS system Coile and Levine (2006) focus on a ggregate t rends i n labor supply r ather t han t he wealth eff ects on i ndividual retirements a nd fi nd no evidence that stock market changes were the driving force in labor supply because, fi rst, few households have substantial stock holdings, and second, they must be extremely respon-sive to market fl uctuations to generate the observed aggregate employment reversal in the recession (i.e., an increase in the labor force participation rate for older workers aged 55–64 between 2000 and 2002)

By contrast, some other studies do fi nd evidence that stock market fl tuations alter retirement behavior Coronado and Perozek (2003) study the impact of the stock market boom on retirement decisions, explaining the diff erence be tween ac tual a nd ex pected r etirement a ges Th ey fi nd that HRS respondents who held equity prior to the bull market of the 1990s retired, on average, 7 months earlier than other respondents when the mar-ket values increased Hermes and Ghilarducci (2006), using current popu-lation survey data, show that the 40% decline in the S&P500 since January

uc-2000 caused the labor force participation of older workers aged 55–64 to increase by 2.64% and 5.36% for men and women, respectively Th ei r sepa-rate estimate on HRS data shows that the probability of retirement for men aged 61–64 with DC plans fell by 10.7 percentage points from 1998 to 2002 However, their results are sensitive to the age ranges selected.*

* Th e magnitude and sign of the interaction term between DC coverage and year 2002 in their probit model is conditional on t he independent variables See Ai and Norton (2003) for a discussion of the econometrics.

Retirement Decision ◾ 293

Another important strand of research is devoted to the eff ect of HI on retirement decisions Gustman and Steinmeier (1994), based on the then-modest employer contribution cost to employee HI ($2500 per year before age 65), fi nd a small eff ect of employer-provided HI on retirement behavior Rust and Phelan (1997), explicitly modeling individual risk aversion and a distribution of health care expenditures in a dynamic lifecycle framework,

fi nd strong impacts of HI and Medicare on retirement, that is, a signifi cant fraction of “HI-constrained” individuals “optimally” remain employed to attain HI coverage until they are eligible for Medicare coverage at age 65 Blau and Gilleskie (2001) show that the availability of employer-provided RHI increases the rate of labor exit Blau and Gilleskie (2008) similarly show that the access and restrictions to RHI and Medicare have a mod-est impact on employment behavior French and Jones (2004) argue that the value of employer-provided HI not only lies in cost reduction but also

in uncertainty reduction for employees Th eir simulations project that a rise in Medicare eligibility age will signifi cantly delay retirement, if work-ers have no other source of insurance but that tied to employment Rust (2005) simulates faculty retirement decisions and shows that an elimina-tion of the retiree health plan (or a substantial reduction of its generosity)

as a cost -cutting measure may signifi cantly reduce the incentive for the existing faculty to retire Mulvey and Nyce (2005) show that, besides DB pension plans, the availability of RHI boosts the likelihood of early retire-ment and that linking the employer-paid insurance premium to service tenure would mitigate such early exits

13.3 DATA

In this analysis, we use the longitudinal, cross-section data from the HRS waves 1992–2004 Th e data set is representative of t he national popula-tion of older workers and retirees and provides detailed information on demographics, health status and insurance coverage, income and wealth, and employment or retirement st atus about A mericans over t he a ge of 50.* Th e respondents and their spouses are (re)interviewed every 2 years

We ex clude t hose o bservations t hat lack a w ork–retirement t ransition Specifi cally, the “AHEAD” cohort respondents (born in 1924 or before) and early baby boomers (born in 1948–1953) are dropped because the for-mer were generally already in the retirement phase when surveyed while the latter were added to the HRS survey in 2004 and therefore have only

* Source: http://hrsonline.isr.umich.edu Th e data and documentation are the Rand version G.

294 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling

one observation We focus on t he retirement behavior of older workers

in t he private sec tor a nd exclude government workers (determined, for each worker, by the job with longest tenure) Th e analysis is focused on the retirement decision for those respondents aged between 50 and 75

Th e fi nal data set consists of the following cohorts: HRS (born in 1931–1941) with survey data 1992–2004, Children of Depression (born in 1924–1930) with survey data 1998–2004, and War Babies (born in 1942–1947) with survey data 1998–2004 Each respondent therefore has up to seven observations An implicit assumption here is that retirement is reversible–survey respondents may return to work (although not necessarily with the same fi rm) aft er retirement Th is is not a stringent assumption given the documented fi ndings of retirement reversal in the literature Ruhm (1990) and Maestas (2007) each fi nd t hat about a q uarter of retirees moved to

“unretirement,” while many others reversed from full retirement to tial r etirement C han a nd S tevens (2008) sh ow t hat abo ut o ne-third o f older individuals who are ever partially or fully retired in the HRS data have reversed their retirement status

par-Retirement is the result of a complex decision-making process Various economic and demographic factors are expected to jointly infl uence the timing and the extent of the transition from employment to retirement,

or a reversal when applicable We run probit regressions to identify factors that may help explain the probability of retirement at any age As retire-ment behavior is person- or household-specifi c, demographic character-istics naturally play an important role In the empirical study, we control for such fac tors as age, gender, educational attainment level, marital sta-tus, spo use em ployment st atus, g ood o r bad h ealth, sel f-employment, occupation, union membership, longevity expectation, and employment–retirement transition through a bridge job Particularly interesting are the impacts of the following broader environmental factors: retirement plan coverage, vulnerability of DC plan accounts to the business cycle, wealth adequacy, earning prospects, SS rules, and HI coverage

Th e appendix de scribes t he construction of a ll va riables u sed i n t he regression a nalysis Table 1 3.1 su mmarizes t he ba sic st atistics o f H RS respondents who are included in the regressions Th ese demographic and

fi nancial statistics indicate that the data sample is fairly representative of the population of older workers Th e H RS respondent-level weights a re used in the regressions All wealth and income values are in constant 2004 dollars and the unit of measurement is $10,000

Retirement Decision ◾ 295

TABLE 13.1 Summary Data Statistics

Variable Median Mean Deviation Standard Minimum Maximum

pension wealth 14.05 23.82 34.11 0 1485.5

SS + DB wealth 21.32 28.66 28.87 0 1521.1

(continued)

296 ◾ Pension Fund Risk Management: Financial and Actuarial Modeling

TABLE 13.1 (continued) Summary Data Statistics

Variable Median Mean Deviation Standard Minimum Maximum

DC plan wealth 1.50 8.00 22.84 0 2163.9 Net housing equity 8.23 12.15 29.96 −638.4 2280.7 Nonhousing fi nancial

wealth 1.88 10.02 42.55 −91.9 3675.5Total household

wealth 41.91 58.84 77.43 −617.8 4791.7Earnings prospect or

opportunity cost 3.21 4.17 5.56 0 404.6Total household

income 6.03 8.16 10.68 0 593.0

DC wealth-earnings

replacement rate (%) 2.92 16.02 43.58 0 400.0

Source: Authors’ calculations based on HRS 1992–2004 survey data.

Note: Wealth and income variables are in constant 2004 terms and the unit is $10,000.

13.4 EMPIRICAL RESULTS: ESTIMATING AND

EXPLAINING THE PROBABILITY OF RETIREMENT

In the probit regression model, the binary dependent variable takes the value of 0 or 1, indicating that the survey respondent is “not retired” or

“retired,” respectively, if the HRS labor force participation data indicates

so We group partial and full retirements together in the main regression, but also treat them separately in an ordered probit model; the results are not sensitive to the alternative specifi cation (see Section 13.4.8) Respondents are also classifi ed as retired if they are older than 65 but the labor force participation status is missing, or if they are not in labor force and older than 62 Observations are dropped if respondents are disabled or if they are not in labor force when younger than 62 Th e variable value of 0 other-wise indicates full- or part-time employment including those unemployed but looking for a f ull time job.* Th ese steps a im to exclude people who have never worked or are only weakly attached to the labor market so as to allow a valid employment–retirement transition Th e regressors are those relevant variables listed earlier Table 13.2 reports the regression results for

* Th e separation of partial employment from partial retirement is directly based on the HRS defi nition: “If he/she is working part-time and mentions retirement,” the labor force status of the survey respondent “is set to partly retired” and “if there is no mention of retirement,” the

status “is set to work ing part-time.” Source: RAND HRS Data Documentation, Version G,

March 2007, p 1035.

Specifi cation 2 DB/DC Coverage

by Wealth Dominance

Specifi cation 3 Wealth Adjusted for Married Couples

Specifi cation 4 Wealth Adjusted and Spousal Observations Excluded

Health insurance, unconditional on

(continued)

Specifi cation 2 DB/DC Coverage

by Wealth Dominance

Specifi cation 3 Wealth Adjusted for Married Couples

Specifi cation 4 Wealth Adjusted and Spousal Observations Excluded

Source: Authors’ regression results based on HRS 1992–2004 survey data.

Notes: Wealth and income variables are in constant 2004 terms and the unit is $10,000 Th e dependent variable in Specifi cations 1–6 takes value 1 = retired

and 0 = not retired It takes value 2 = fully retired, 1 = partially retired, and 0 = not retired in Specifi cation 7, where results shown are coeffi cients, not marginal eff ects Th e HRS respondent-level weights are used in the probit regressions.

*, **, and *** denote signifi cance at the 0.10, 0.05, and 0.01 level, respectively; insignifi cant otherwise.