You are on page 1of 17

J Labor Res (2015) 36:274290

DOI 10.1007/s12122-015-9207-y

Unemployment and the Retirement Decisions of Older


Workers
Paul Marmora Moritz Ritter

Published online: 31 March 2015


Springer Science+Business Media New York 2015

Abstract This paper examines how unemployment late in workers careers affects
retirement timing. Using data from the Survey of Income and Program Participation from 1996 to 2011, we document that unemployed workers permanently leave
the labor force at a significantly higher rate than employed workers. This effect is
stronger once workers become eligible for Social Security benefits. The effect of
unemployment on retirement early in an unemployment spell is weaker for workers
eligible for UI benefits. Unemployed workers, particularly those workers in households with below median wealth, also have a significantly higher rate of early Social
Security uptake shortly after turning 62 relative to employed workers.
Keywords Older workers Retirement Social security Unemployment
duration Unemployment benefits
JEL Classification J26 H55 J14 J64 J65

Introduction
Older workers face a lower risk of job loss than do younger workers. But when
they lose their jobs, older workers have longer unemployment spells (Farber 2005;

P. Marmora
Department of Economics and Business, RandolphMacon College, Ashland, VA 23005, USA
e-mail: paul.marmora@gmail.com
M. Ritter ()
Department of Economics, Temple University, 1301 Cecil B. Moore Ave, Philadelphia, PA 19122,
USA
e-mail: moritz.ritter@temple.edu

J Labor Res (2015) 36:274290

275

Johnson 2005).1 Long unemployment spells are of particular concern, since the risk
of workers becoming discouraged and less connected to the labor market increases
with the length of the unemployment spell. This issue has garnered more attention
in the aftermath of the great recession, as unemployment remained elevated for an
extended time and labor force participation rate for all age groups declined in recent
years (e.g. Fujita 2013, Kroft et al. 2014, Krueger et al. 2014). This paper studies the
role of unemployment duration and the availability of unemployment benefits on the
decision of older workers to leave the labor market permanently.
The topic of long term unemployment and labor market participation of older
workers is of particular interest for three reasons. First, older workers make up a
disproportionate number of long term unemployed.2 Second, after turning 62, older
workers typically become eligible for Social Security benefits, which may facilitate
leaving the labor force relative to individuals with no income substitute available.
Any early uptake of Social Security benefits results in reduced monthly payments
to the retiree for the remainder of her life, which represents a significant effect on
the workers welfare. Last, policymakers need a better understanding of employment
uncertainty for older workers when considering reforms to the Social Security system
(Bentez-Silva et al. 2012, 2014).
In this paper, we use individual-level panel data from the Survey of Income and
Program Participation (SIPP) to study how unemployment affects workers retirement decision. The SIPP provides monthly information on the exact timing of
individuals labor market transitions, as well as detailed information on their income
sources. In particular, the SIPP provides information on Social Security and unemployment insurance (UI) benefits reiceipt, which allows us to determine whether
the retirement decision was preceded by an unemployment spell, how long that
unemployment spell lasted, and whether the worker received UI benefits.
We document that job loss and the subsequent unemployment spell have a large
effect on the timing of permanent labor force exit (retirement) and Social Security
uptake for older workers. In particular, we find that workers who are not receiving
UI benefits tend to retire quickly after a job loss. The effect is significantly larger for
workers eligible for early Social Security benefits. For workers receiving UI benefits,
we find a spike in retirement shortly after UI benefits expire, with said spike significantly larger for workers eligible for early Social Security benefits. Moreover, we
find that unemployment on ones 62nd birthday increases the probability of Social
Security uptake within six months of ones 62nd birthday by 46.1 percentage points
for individuals with below median household wealth, but only by 12.0 percentage
points for individuals with above median household wealth.

1 At

the end of 2014, the U.S. unemployment rate among workers 55 through 64 years of age was 3.8 %,
well below the current national average of 5.6 %. However, the average and median unemployment duration for workers 55 to 64 years of age was 54.6 and 20.5 weeks, respectively, compared to an overall
average and median duration of 32.4 and 13.0 weeks, respectively (BLS).
2 In 2013, 30 % of long-term unemployed were over 50 years of age, while the same age group only made
up 20 % of short-term unemployed (Krueger et al. 2014).

276

J Labor Res (2015) 36:274290

One possible explanation for these findings is that unemployed workers may have
to rely on Social Security benefits because they cannot find new employment before
running out of UI benefits (income hypothesis). While an alternative simple selection
hypothesis (workers who are more likely to retire early are also more likely to lose
their jobs) is equally consistent with the basic finding that the retirement rate for
unemployed workers is higher, the dependence of the retirement decision on Social
Security eligibility, UI coverage and wealth all suggest that income replacement plays
a significant role in the retirement decision.
Moreover, during times of when Congress extended UI benefits to 99 weeks,
we continue to find that UI covered workers retire at higher rates right after their
(extended) coverage expires, while non-UI covered workers retire shortly into their
unemployment spell. Since the selection effect is expected to be less strong during
times of high unemployment, this finding supports the income hypothesis. Using
information on health insurance coverage, we also find that unemployed workers
who are not covered by health insurance are less likely to retire than covered workers, again indicating that pecuniary aspects play a significant role in the retirement
decision.
Related Literature
We are not the first to document that unemployment late in a workers career has a
significant effect on the retirement decision. Using data from the Health and Retirement Study (HRS), Chan and Stevens (1999, 2001, 2004) find that displaced workers
have significantly lower future employment probabilities and retire at substantially
higher rates than non-displaced workers. The authors at least partially attribute this
finding to workers finding it difficult to find reemployment.3 Similarly, Coile and
Levine (2007, 2011a, b) use the Current Population Survey (CPS) to show that
retirement rates increase during weak labor markets and more so for less educated
workers, suggesting that Social Security may serve as a form of unemployment insurance for these workers. Both the HRS and the CPS have the advantage of being
larger and providing a longer panel than the SIPP, respectively, but do not provide detailed and high frequency information on the unemployment spell duration
prior to retirement. An advantage of the SIPP is that it allows us to study the role
of unemployment duration in the retirement decision. Our exercise contributes to
the literature by providing more direct evidence on the hypothesis that older job
seekers might claim Social Security benefits early in an effort to mitigate income
losses.
Our findings are also consistent with results by Tatsiramos (2010) and GarciaPerez et al. (2013). Tatsiramos (2010) uses data from four European countries and
finds that unemployed workers in Germany and Spain (countries with generous
unemployment insurance for older workers and generous early retirement provisions)
have significantly lower reemployment rates than older workers an Italy and the UK.

3 Also using the HRS, Goda et al. (2011) and Gustman et al. (2011) study the wealth effect from asset
market fluctuations on the retirement decision.

J Labor Res (2015) 36:274290

277

Garcia-Perez et al. (2013) confirm the importance of the institutional features in the
case of Spain using administrative data. They find that unemployed workers who
exhaust their unemployment benefits retire at high rates.

Data
We use data from the 1996, 2001, 2004, and 2008 panels of the Survey of Income and
Program Participation (SIPP). Each SIPP panel follows the same individuals over a
period ranging from 48 months in the 1996 panel to 32 months in the 2008 panel.
For each individual, the SIPP provides employment status at a weekly frequency and
income at a monthly frequency, as well as information on total household wealth,
typically at an annual frequency.4 The relatively high frequency of labor force information allows us to determine whether the retirement decision was preceded by an
unemployment spell, and how long any unemployment spell lasted. We restrict our
sample to individuals age 55-69 who reported at least one month of labor force participation during the sample period.5 This results in a sample with 38,998 individuals
and 894,687 monthly observations for which an individual was in the labor force
during the previous month.
Retirement and Unemployment
For the purposes of our analysis, we require that two conditions be met in order
to classify an individual as retired. First, we follow the literature (e.g. Coile and
Levine 2011a) and define retirement as complete labor force withdrawal: a person
exits the labor force for the rest of the survey in month t after participating in the
labor force in teh prior month, t 1. Second, to account for temporary labor force
exits for which we do not observe the end of the spell, we require an individual to
self-report that they are either out of the labor force because they are retired, or that
they have retired from a job at least once during their lifetime. Using this definition
of retirement, we find that the average monthly retirement rate for workers aged 55
to 69 from 1996 to 2011 is 0.68 % (7.9 % annually), which is only slightly below the
9 % annual retirement rate in the March CPS from 1980 to 2007 reported in Coile
and Levine (2011a).
We consider an individual to be unemployed if they are without a job and report
their labor market status as no job/business-looking for work and we define a job
loss as a separation into unemployment. In other words, if a worker is holding a
job in t and reports being unemployed in t + 1, we consider her as having lost

4 Respondents are interviewed three times per year about their experiences over the previous fourth months.

Some core information is collected at every interview, while other topical information is collected less
frequently.
5 We do not restrict the sample further by gender or race but require the topical asset module to have been
answered at least once. The results are qualitatively and quantitatively similar for a male only sample.

278

J Labor Res (2015) 36:274290

Table 1 Summary statistics by SIPP panel


Panel

Individuals

Monthly

Retirements

Job Losses

Observations

Retirements
while Unemployed

1996

8,167

211,694

1,802

463

132

2001

7,198

156,016

1,184

437

128

2004

11,581

273,630

1,717

815

310

2008

12,052

253,347

1,518

1,006

295

her job.6 Table 1 lists the number of observations and instances of retirement, job
loss and retirement while unemployed for each panel of the SIPP and Fig. 1 shows
the difference in monthly retirement hazards between the employed and the unemployed, by age. Predictably, the monthly retirement rate for unemployed respondents
(3.2 %) is much higher than the monthly retirement rate for the employed (0.61 %).
Unemployed workers also experience much larger spikes in retirement around the
eligibility ages for pension and social security benefits: the monthly retirement rate
for unemployed individuals age 62 and older is 7.0 %, whereas the same rate for
employed individuals is only 1.2 %, which suggests that unemployed workers might
draw on social security and their retirement savings as alternative sources of income.
Retirement and Total Household Wealth
SIPP respondents are asked detailed questions on wealth in the Assets and Liabilities
topical module questionnaire, typically administered once a year. We focus on the
measure of total net worth at the household level and we assign individuals to quartiles based on total net worth at time the questionnaire is administered for the first
time.7 In order to estimate the fixed effect model by wealth quartile, the individuals
wealth quartile remains constant throughout the panel.8
Table 2 lists the retirement, job loss and unemployment rates for each wealth quartile. Not controlling for age and labor market status, individuals in the lowest wealth
quartile retire only at a slightly lower rate (7.2 % annually) than individuals in the
other three quartiles (7.98.1 % annually). On the other hand, the risk of job loss into
unemployment decreases strongly with wealth, from 0.5 % monthly separation rate
for the lowest wealth quartile, to 0.2 % for the highest quartile.
6 While

the SIPP includes a question on the reason for job loss, the response is missing for about one third
of all job losses. Restricting the sample to only those who responded having had an involuntary job loss
renders the sample too small for the detailed duration analysis. However, pooling unemployment duration
leads to qualitatively similar results between both samples; the monthly retirement hazard for those eligible
for early social security benefits is 3.9 percentage points following an involuntary separation, rather than
5.0 percentage points in the sample including all job losses.
7 The primary reason for grouping into quartiles based on net worth instead of using the raw measure of
wealth is that total net worth in the SIPP supplemental questionnaire is underreported, particularly for
wealthy households (Czajka et al. 2003).
8 31.2 % of the individuals in our sample switch wealth quartiles during the interview period. Reassigning
these individuals upon switch does not affect the results significantly.

279

.02

Hazard
.04

.06

.08

J Labor Res (2015) 36:274290

54

56

58

60

62
Age

Employed

64

66

68

70

Unemployed

Fig. 1 Average monthly retirement rate by labor force status

Figure 2 shows the age profile of retirement rates for those below and those above
the median total household wealth, by age. Higher-asset individuals retire at a slightly
higher rate at younger ages, but the trend switches exactly at age 62 and spikes in
retirement at age 62 and 65 are considerably more pronounced for those below the
median wealth.9 Since 62 and 65/66 are the early and full retirement ages, respectively, this pattern is consistent with individuals who have few financial assets making
their retirement decision based on social security availability.10 On the other hand,
wealthier workers can afford to make a more independent decision, giving rise to the
hypothesis that an unexpected shock to the stock or housing markets may induce high
wealth individuals to postpone retirement.

Empirical Analysis of the Retirement Decision


Income Hypothesis and UI Recipiency
The previous section demonstrated that both unemployment and wealth play a role in
an individuals retirement decision. In this section, we seek to quantify these effects

9 The

difference in retirement at age 62 is statistically significant at the 10 % level.


(normal) retirement age increased during the sample from 65 if born in 1937 or earlier to 66 if born
between 1943 and 1954.
10 Full

280

J Labor Res (2015) 36:274290

Table 2 Employment and retirement by total household wealth


Full Sample 1st Quartile 2nd Quartile 3rd Quartile 4th Quartile
Monthly Retirement Rate

0.0068

0.0062

0.0068

0.0070

0.0069

Unemployment Rate

0.029

0.053

0.033

0.024

0.016

Monthly Job Loss Rate

0.0029

0.0053

0.0032

0.0025

0.0017

Fraction Female

0.470

0.501

0.498

0.472

0.430

Fraction White

0.803

0.605

0.753

0.851

0.907

Number of Individuals

38,998

7,527

9,511

10,472

11,488

162,430

214,557

244,117

273,583

into Unemployment

Number of Monthly Observations 894,687

and explore how the unemployment effect varies with unemployment duration, the
eligibility for Social Security benefits, and the individuals household wealth. We use
a difference-in-differences linear-probability-model approach to identify the effects
of labor market conditions on retirement behavior. Exploiting the panel nature of the
SIPP, we estimate a fixed effects model, which allows us to control for unobserved
heterogeneity that could potentially drive our results (for example, workers who are
more likely to be unemployed are also more likely to leave the labor force). Thus,
our main specification is
Retireit = 0 + 1 U nemployed Groupit Age Groupit + 3 Xit + i + it (1)
where Retireit is an indicator variable, equal to one for a worker i being in the labor
force in month t but not month t + 1 or any other future month. The key explanatory variables of interest are the indicator variables for the six UI recipiency/duration
groups (described below), while allowing their effects to vary with Social Security
eligibility.11
To study the effect of the expiration of UI benefits on retirement, we first classify all unemployment spells in our sample into two groups: spells for which the
unemployed individual received benefits at least once and those spells for which no
benefits were ever reported.12 Next, we break up each unemployment spell into three
sub-periods: the initial period after job loss, with potential UI coverage (typically

11 We also include a quadratic polynomial in age and indicator variables for the person being of early
(EarlySSage) and full Social Security (F ullSSage) eligibility age. In addition, we include state and time
fixed effects and an indicator variable for the nth observation in each panel, which controls for the fact
that retirements are more likely to occur later in the panel due to the definition of retirement as exiting the
labor force for the rest of the survey. We also include an indicator variable for the start of a new wave to
address potential seam-bias problems.
12 There are two potential concerns in proxying for UI eligibility in this way. First, takeup of government
programs is known to be undereported in surveys (e.g. Meyer et al. 2009). The underreporting would cause
a downward bias in our estimates. Second, takeup is not random; if individuals with lower job-finding
probability and a greater income need are more likely to take up UI, this would bias our estimates upward.

281

.005

Hazard
.01

.015

.02

J Labor Res (2015) 36:274290

54

56

58

60

62
Age

Wealth Above Median

64

66

68

70

Wealth Below Median

Fig. 2 Average monthly retirement rate by household wealth

26 weeks, but may be longer during times of extended benefits due to a weak labor
market); the 26 weeks after the expiration of potential UI benefits (after period)
and more than 26 weeks after the expiration date (long term period). Thus, at any
point during an unemployment spell, an individual belongs to one of six groups,
determined by UI recipiency and time since job loss (the three sub-periods).
The results are reported in Table 3. For all age groups, unemployed workers not
receiving UI benefits have a significantly larger retirement probability during the UI
coverage period than do observationally equivalent employed workers. Workers not
eligible for UI benefits retire early in their unemployment spell. The effect of unemployment is significantly larger once workers reach Social Security eligibility age.
Before age 62, being unemployed and not receiving UI benefits increases a workers
monthly retirement rate by 1.6 percentage points; after turning 62, the equivalent
effect is 7.0 percentage points.
Receiving UI benefits significantly dampens the effect of unemployment early in
the spell: before age 62, the retirement rate of unemployed workers receiving UI
benefits is statistically indistinguishable from the retirement rate for employed workers. After age 62, the effect of unemployment falls to 2.8 percentage points with
the receipt of UI benefits. After the eligibility for UI benefits runs out, but before
age 62, workers previously receiving benefits behave similarly to those who have
never received benefits. However, after age 62, workers who lost their UI benefits

282

J Labor Res (2015) 36:274290

Table 3 Retirement transitions by unemployment spell length


Full Sample

Below Median Above Median


Wealth

Wealth

Mean of Dependent Variable

0.0068

0.0066

0.0069

EarlySSage

0.0058***
(0.0008)
0.0111***
(0.0014)

0.0061***
(0.0012)
0.0121***
(0.0023)

0.0057***
(0.0010)
0.0106***
(0.0019)

0.0164***

0.0150***

0.0182***

(0.0022)

(0.0027)

F ullSSAge
Age 55 61*
U nemployed Bef ore Expiration

U nemployed Bef ore Expiration*U I Recipiency 0.0136*** 0.0139***

(0.0039)
0.0135***

(0.0025)

(0.0030)

(0.0042)

0.0120*

0.0037

0.0354***

(0.0064)

(0.0065)

(0.0124)

0.0051

0.0189*

0.0169

(0.0089)

(0.0098)

(0.0165)

0.0264*

0.0342

0.0122**

(0.0148)

(0.0249)

(0.0054)

0.0097

0.0104

0.0422

(0.0219)

(0.0305)

(0.0310)

0.0655***
(0.0097)
0.0389***
(0.0117)
0.0324
(0.0297)

0.0743***
(0.0119)
0.0452***
(0.0142)
0.0111
(0.0113)

0.1131***

0.0901*

0.1399*

(0.0413)

(0.0501)

(0.0724)

U nemployed Long T erm

0.0026

0.0040

0.0079

(0.0048)

(0.0109)

(0.0074)

U nemployed Long T erm*U I Recipiency

0.0378***

0.0137

0.0368*

(0.0102)

(0.0152)

(0.0189)

0.0588***

0.0489***

0.0712***

(0.0113)

(0.0148)

U nemployed Af ter Expiration


U nemployed Af ter Expiration*U I Recipiency
U nemployed Long T erm
U nemployed Long T erm*U I Recipiency
EarlySSage*

0.0697***
(0.0076)
U nemployed Bef ore Expiration*U I Recipiency 0.0418***
(0.0092)
U nemployed Af ter Expiration
0.0215
(0.0172)
U nemployed Bef ore Expiration

U nemployed Af ter Expiration*U I Recipiency

F ullSSage*
U nemployed Bef ore Expiration

U nemployed Bef ore Expiration*U I Recipiency 0.0364*** 0.0393**


U nemployed Af ter Expiration
U nemployed Af ter Expiration*U I Recipiency

(0.0176)
0.0342

(0.0130)

(0.0161)

(0.0213)

0.2393***

0.1714

0.2925***

(0.0636)

(0.1100)

(0.0435)

0.2289*** 0.1590

0.2950***

(0.0659)

(0.0441)

(0.1120)

J Labor Res (2015) 36:274290


Table 3

283

(continued)
Full Sample

Below Median

Above Median

Wealth

Wealth

0.0885

0.0842

(0.0642)

(0.0638)

Number of Observations

894687

376987

517700

Number of Individuals

38998

17038

21960

U nemployed Long T erm


U nemployed Long T erm*U I Recipiency

s.e. in parentheses
(***) denotes significance at the 1 % level, (**) at the 5 % level, (*) at the 10 % level
Each specification is a fixed effects regression with Retire as the dependent variable. We also include
controls for age and age-squared, as well as time and state fixed effects, an indicator variable for the
beginning of a new wave, and an indicator variable for the nth observation in each panel. Standard errors
are clustered at the individual level

experience a large spike in retirement. Workers eligible for early Social Security and
who ran out of UI benefits retire at a rate 11.3 percentage points higher than those
never eligible for UI benefits and 13.5 percentage points higher than employed workers eligible for early Social Security benefits. The differential retirement behavior
of workers receiving by and UI benefits and elibible for Social Security benefits is
consistent with the interpretation that unemployed workers use retirement benefits to
help make up the income loss associated with their job loss.
Robustness: 99 Week Extension
To explore whether workers retirement is driven by discouragement rather than
income replacement, we take advantage of our samples coverage of the recent 99
week extension of UI eligibility. If retirement is driven by the income gap once eligibility for UI benefits expires, we would expect to find a strong effect following the
expiration of regular benefits, at 26 weeks, but would not expect a comparable effect
during any extension. In other words if retirement is driven by income replacement
rather than a simple discouragement effect, we expect a covered worker to respond
differently in week 27 of his unemployment spell if the coverage period is 99 weeks
than if it is only 26 weeks.13
All unemployment spells are divided at 26 weeks, regardless of whether benefits
are extended or not. We then include the indicator for UI benefit recipiency and an
indicator for the 99 week extensions. As shown in Table 4, we find that UI covered
workers retire at a lower rate during periods with regular benefits. Since the sample

13 We omit periods with extensions shorter than 99 weeks to focus on the longest possible UI duration and
a clear separation relative to the baseline of 26 weeks.

284

J Labor Res (2015) 36:274290

Table 4 Effect of UI extension on retirement transitions


Full Sample
Mean of Dependent Variable

0.0069

EarlySSage

0.0070***
(0.0010)

F ullSSage

0.0137***
(0.0018)

U nemployed 1 26 W eeks

0.0265***

U nemployed 1 26 W eeks*U I Recipiency

0.0194***

(0.0034)
(0.0040)
U nemployed Af ter 26 W eeks

0.0216***

U nemployed Af ter 26 W eeks*U I Recipiency

0.0050

(0.0070)
(0.0099)
99Extension*
U nemployed 1 26 W eeks

0.0011
(0.0073)

U nemployed 1 26 W eeks*U I Recipiency

0.0046
(0.0080)

U nemployed Af ter 26 W eeks

0.0162

U nemployed Af ter 26 W eeks*U I Recipiency

0.0315**

Number of Observations

650023

Number of Individuals

35992

(0.0120)
(0.0144)

s.e. in parentheses
(***) denotes significance at the 1 % level, (**) at the 5 % level, (*) at the 10 % level
The specification is a fixed effects regression with Retire as the dependent variable. We also include
controls for age and age-squared, as well as time and state fixed effects, an indicator variable for the
beginning of a new wave, and an indicator variable for the nth observation in each panel. Standard errors
are clustered at the individual level

is pooled by age, however, we no longer observe a large spike around expiration of


benefits as we do for the covered group eligible for early Social Security benefits in
Table 3.
The bottom half of Table 4 shows the differential retirement rates during periods of
extended benefits. In the first 26 weeks of unemployment, workers retire at the same
rate as during normal times. However, after 26 weeks, UI covered workers retire at
significantly lower rates than do UI covered workers whose coverage expired during

J Labor Res (2015) 36:274290

285

normal times. After 26 weeks, the monthly retirement rate is 3.2 percentage points
lower than the rate for comparable workers who run out of benefits during regular
benefit periods. On the other hand, workers not receiving UI benefits exhibit retirement behavior no different than that of workers not covered during regular benefit
periods. To the extent that unemployment was less subject to selection during the
Great Recession than during normal times, we also view this as support of the income
hypothesis and not the selection hypothesis.
Other Factors in the Retirement Decision
The decision to retire is a complex one and unemployment and earnings loss are but
two factors in this decision. An additional factor that has been shown to affect the
retirement decision is the availability of health insurance, which typically is employer
provided in the U.S. (e.g. Rogowski and Karoly 2000). Having health insurance (e.g.
through a spouse) may make it easier to leave the labor force upon job-loss.
We use information on health insurance coverage in the SIPP and replicate the
analysis from the previous section, including an indicator for no private health insurance coverage.14 As Table 5 shows, workers without employer sponsored health
insurance are more likely to leave the labor force at all ages. However, workers without health insurance are considerably less responsive to unemployment. Before age
62, having no health insurance fully offsets the effect of unemployment on retirement. From age 62 to the full Social Security age, the effect of unemployment on
uncovered workers is about half the effect of unemployment of covered workers. Yet,
unemployed workers without health insurance coverage after reaching the full Social
Security age (and becoming eligible for Medicare) retire at the same rate as workers
with health insurance.
Taken together, these findings indicate that workers between the ages of 62 and
full Social Security age who do not have private health insurance are likely seeking employment providing insurance. Considering the substantial monetary value
an employer provided health plan can represent, we view this finding as additional
support for the interpretation that the retirement decision of unemployed workers is
heavily driven by financial consideration.
An additional consideration is retirement incomes that follow rules different from
Social Security, most notably defined benefits pensions. In our sample, there are only
122 job losses for workers who are covered by a defined benefit pension. Therefore, we cannot replicate our analysis above while separately accounting for defined
benefits pension. However, we can say that workers older than age 60 (age 62) who
are covered by a defined benefits pension have a higher probability of finding new

14 Since the resulting table becomes very difficult to read due to the multitude of interaction terms and since
accounting for health coverage does not change our results qualitatively, we present a simple regression
table that excludes the duration controls to illustrate the main findings.

286

J Labor Res (2015) 36:274290

Table 5 Retirement transitions, FE


(1)

(2)

(3)

Mean of Dependent Variable

0.0068

0.0068

0.0068

EarlySSage

0.0057***

0.0051***

0.0052***

(0.0008)

(0.0008)

(0.0008)

0.0109***

0.0105***

0.0107***

F ullSSage
U nemployed

(0.0014)

(0.0015)

(0.0015)

0.0106***

0.0153***

0.0141***

(0.0015)
no H ealth I ns
U nemployed*EarlySSage
U nemployed*F ullSSage

(0.0022)

(0.0020)

0.0061***

0.0048***

(0.0009)

(0.0008)

0.0389***

0.0443***

0.0449***

(0.0057)

(0.0075)

(0.0074)

0.0325***

0.0320***

0.0384***

(0.0081)

(0.0103)

(0.0102)

0.0157***

0.0108***

(0.0035)

(0.0028)

0.0240**

0.0308**

(0.0122)

(0.0125)

U nemployed*no H ealth I ns
U nemployed*no H ealth I ns*EarlySSage
U nemployed*no H ealth I ns*F ullSSage

0.0008

0.0233

(0.0161)

(0.0165)

Number of Observations

894687

894687

894687

Number of Individuals

38998

38998

38998

s.e. in parentheses
(***) denotes significance at the 1 % level, (**) at the 5 % level, (*) at the 10 % level
Each specification is a fixed effects regression with Retire as the dependent variable. We also include
controls for age, house prices and interaction terms of each variable with the sate unemployment rate
(not significant), as well as time and state fixed effects, an indicator variable for the beginning of a new
wave, and an indicator variable for the nth observation in each panel. Standard errors are clustered at the
individual level
Specification (2) defines no H ealth I ns as no private health coverage and specification (3) defines
no H ealth I ns as no health coverage (including no Medicaid)

employment than other workers: 69.6 % vs. 55.4 % (63.2 % vs. 51.1 % at age 62).
However, these differences are not statistically significant.
Income Hypothesis and Social Security Eligibility
Another prediction of the income hypothesis is a strong effect of unemployment on
retirement rates at the time a worker becomes eligible for Social Security benefits

J Labor Res (2015) 36:274290

287

Table 6 Retirement at 62nd Birthday


Full Sample

Below Median

Above Median

Wealth

Wealth

Mean of Dependent Variable

0.0923

0.0959

0.0899

U nemployed

0.1398***

0.2238***

0.0637

(0.0507)

(0.0810)

(0.0578)

0.0087

0.0109

0.0100

(0.0082)

(0.0137)

(0.0106)

0.0031

0.0079

0.0073

(0.0161)

(0.0249)

(0.0184)

4276

1779

2497

U ERate
U nemployed*U ERate
Number of Individuals
s.e. in parentheses

(***) denotes significance at the 1 % level, (**) at the 5 % level, (*) at the 10 % level
Each specification is a OLS regression with Retire W ithin 6 Months as the dependent variable. We
also include time and state fixed effects, an indicator variable for the beginning of a new wave, and an
indicator variable for the nth observation in each panel

at age 62. We focus therefore on the first 6 months after an individual turns 62 and
estimate the effect of unemployment in the month of a workers 62nd birthday on the
probability of retirement (as defined previously) and actual Social Security uptake
within 6 months of her 62nd birthday.15
Overall, about 9.2 % of workers in our sample who are in the labor force on their
62nd birthday leave the labor force within 6 months. As shown in Table 6, being
unemployed on ones 62nd birthday has a large and significant effect on the 6 month
retirement probability: for individuals with below median wealth, all else equal,
being unemployed increases the probability of a retirement shortly after becoming
eligible for Social Security benefits by 22.4 percentage points. For individuals with
above median wealth, the effect is only 6.4 percentage points and not statistically
significant.
Our definition of retirement above focuses on labor force exit, rather than Social
Security uptake. The main reason for this is the lapse of time between the application
for Social Security benefits and the beginning of payouts, which complicates the
identification of the exact timing of the retirement decision. Regardless, if retirements
by the unemployed are truly driven by the need for an income, as suggested above,
we should see a large effect of unemployment on Social Security uptake. We can
again use the discontinuity in eligibility at age 62 to study the effect of unemployment
on uptake of Social Security benefits. Following the same approach as above, we
estimate the effect of being unemployed on the 62nd birthday on Social Security

15 We require individuals to start reporting income from Social Security within 6 months of the 62nd
birthday. This requires an application for Social Security benefits within 2-3 months of turing 62, at the
latest.

288

J Labor Res (2015) 36:274290

Table 7 Social security withdrawl at 62nd birthday


Full Sample

Below Median

Above Median

Wealth

Wealth

Mean of Dependent Variable

0.2642

0.2939

0.2440

U nemployed

0.2956***

0.4612***

0.1196

(0.0619)

(0.0794)

(0.0758)

0.0003

0.0104

0.0033

(0.0117)

(0.0194)

(0.0155)

0.0002

0.0055

0.0077

(0.0215)

(0.0281)

(0.0339)

4276

1779

2497

U ERate
U nemployed*U ERate
Number of Individuals
s.e. in parentheses

(***) denotes significance at the 1 % level, (**) at the 5 % level, (*) at the 10 % level
Each specification is a OLS regression with Social Security W ithin 6 Months as the dependent variable. We also include time and state fixed effects, an indicator variable for the beginning of a new wave,
and an indicator variable for the nth observation in each panel are also included

uptake within 6 months of turning 62.16 Overall, in our sample of workers in the labor
force, 29.4 % of workers with below median household wealth and 24.4 % of workers
with above median wealth take up Social Security within 6 months of turning 62.
The results are reported in Table 7 and are supportive of the income hypothesis: for individuals with below median assets, being unemployed on the 62nd
birthday increases the probability of taking up Social Security by 46.1 percentage points, compared to only 12.0 percentage points for an individual with above
median wealth. Since the average take-up rate for below median wealth individuals in our sample is 29.4 %, this suggests that the large majority of unemployed
workers with below median wealth take up Social Security early. Also, consistent
with the income hypothesis, we do not find the same effect for above median wealth
individuals.

Discussion
Using data from the SIPP, we document that job loss and the subsequent unemployment spell have a large effect on the timing of retirement and Social Security
uptake. Workers who do not receive UI benefits retire early in their unemployment
spell and the effect is amplified once workers become eligible for early Social Secu-

16 We require individuals to start reporting income from Social Security within 6 months of the 62nd birthday. Due to the aforementioned processing time, this requires an application for Social Security benefits
within 23 months of turing 62, at the latest.

J Labor Res (2015) 36:274290

289

rity benefits. Receiving UI benefits dampens the effect of unemployment early in


the spell but the retirement rate spikes once workers are no longer eligible for UI
benefits.
While our baseline results can also be accounted for by a selection effect, our
complementary results all suggest that unemployment late in a workers career and
the need to find an alternative income source are an important trigger for early retirement. Since retiring early leads to a permanent reduction in the monthly Social
Security benefits an individual will receive for the rest of her life, our findings
suggests that unemployment late in a workers career may have a significant and
lasting effect on that workers remaining-life welfare. Considering the relatively
low job finding rate for older workers and the foreseeable need to reform the
Social Security system, we believe that this issue warrants further attention in future
research.

References
Bentez-Silva H, Garca-Perez JI, Jimenez-Martn S (2012) The effects of employment uncertainty and
wealth shocks on the labor supply and claiming behavior of older American workers. Working Paper.
Bentez-Silva H, Garca-Perez JI, Jimenez-Martn S (2014) Reforming the U.S. Social Security system
accounting for employment uncertainty. Working Paper
Chan S, Stevens AH (1999) Employment and retirement following a late-career job loss. Am Econ Rev
89(2):211216
Chan S, Stevens AH (2001) Job loss and employment patterns of older workers. J Labor Econ 19(2):484
521
Chan S, Stevens AH (2004) How does job loss affect the timing of retirement? B.E. J Econ Anal Policy
3(1):126
Coile CC, Levine PB (2007) Labor market shocks and retirement: do government programs matter? J
Public Econ 91(10):19021919
Coile CC, Levine PB (2011a) The market crash and mass layoffs: how the current economic crisis may
affect retirement. B.E. J Econ Anal Policy 11(1):142
Coile CC, Levine PB (2011b) Recessions, retirement and social security. Am Econ Rev 101(3):23
28
Czajka JL, Jacobson JE, Cody S (2003) Survey estimates of wealth: a comparative analysis and review
of the survey of income and program participation. Mathematica Policy Research, Inc, Washington.
Document No. PR03-45
Farber H (2005) What do we know about Job Loss in the United States? Evidence from the Displaced
Workers Survey, 1981-2004. Economic Perspectives, Federal Reserve Bank of Chicago, QII, pp 13
28
Fujita S (2013) On the causes of declines in the labor force participation rate. Federal Reserve Bank of
Philadelphia, Research Rap
Garcia-Perez JI, Jimenez-Martin S, Sanchez-Martin AR (2013) Retirement incentives, individual heterogeneity and labor transitions of employed and unemployed workers. Labour Econ 20:106
120
Goda GS, Shoven JB, Slavov SN (2011) What explains changes in retirement plans during the great
recession? Am Econ Rev 101(3):2934
Gustman AL, Steinmeier TL, Tabatabai N (2011) How did the recession of 20072009 affect the wealth
and retirement of the near retirement age population in the health and retirement study? NBER
Working Paper No. 17547
Johnson RW (2011) Mommaerts, C, Age differences in job loss, job search, and reemployment. Urban
Institute Discussion Paper
Kroft K, Lange F, Notowidigdo MJ, Katz LF (2014) Long-term unemployment and the great recession:
the role of composition, duration dependence and non-participation. Working Paper

290

J Labor Res (2015) 36:274290

Krueger AB, Cramer J, Cho D (2014) Are the long-term unemployed on the margins of the labor market?
Brookings Papers on Economic Activity
Meyer BD, Mok WKC, Sullivan JX (2009) The under-reporting of transfers in household surveys: its
nature and consequences. NBER Working Paper No. 15181
Rogowski J, Karoly L (2000) Health insurance and retirement behavior: evidence from the health and
retirement survey. J Health Econ 19(4):529539
Tatsiramos K (2010) Job displacement and the transitions to re-employment and early retirement for nonemployed older workers. Eur Econ Rev 54(4):517535

You might also like