Economic American Review: Papers & Proceedings 2011, 101:3, 2328
http://www.aeaweb.org/articles.php?doi=10.1257/aer.101.3.23
Recessions and RetiRement
Recessions, Retirement, and Social Security
By Courtney C. Coile and Phillip B. Levine*
The economic crisis that began in 2008 had
multiple implications for retirement behavior
(Coile and Levine 2007, 2009, and 2010). The
stock market crash may have caused some individuals to defer retirement because of losses in
their 401(k)-type retirement plans. The spike in
unemployment may have led others to retire
sooner in response to a job loss or the inability to find work. By the end of 2010, however,
stocks had nearly rebounded to precrash levels. Although there may have been a short-term
impact, the markets sharp rise has substantially
diminished the importance of this part of the
story. The weakness in the labor market, however, continues to be extensive and persistent.
The purpose of this analysis is to focus on its
implications for retirement and retirement
income in the coming years.
High unemployment has the potential to significantly alter older workers income, both in
the present and for the remainder of their lives.
If an older worker loses his or her job, finding a
new one may be even more difficult than it is for
a younger worker.
Older workers relatively short time horizon
in the labor force may reduce their own or a
prospective employers willingness to invest in
additional human capital. If specific capital has
been lost, the wage a new employer might offer
could be considerably below the workers previous wage, reducing the likelihood of an offer
being extended or accepted. For these reasons
and others, the future employment prospects of
a laid-off older worker may be bleak.
In such an environment, Social Security benefits may be a lifeline for these workers. Taking
up benefits when they first become available at
age 62 may help a laid-off older worker pay the
bills, but it also means a reduced flow of income
for the remainder of the workers life, since
monthly benefits are reduced for early claiming.
The impact of lower monthly payments may be
particularly pronounced for individuals with
low socioeconomic status, who rely heavily on
Social Security to make ends meet.
The purpose of this paper is to examine how
labor market fluctuations affect retirement,
Social Security claiming, and subsequent Social
Security income of older individuals. To this
end, we use 30 years of data from the March
Current Population Survey (CPS) as well as
data from the 2000 Census and 2001, 2002,
and 20062009 American Community Surveys
(ACS). We find support for the notion that workers are more likely to leave the labor force, to
collect Social Security earlier, and to have lower
Social Security income later in life when a
recession hits around the time of retirement. The
impact is greatest for the less-educated, who
are more susceptible to job loss and rely more
heavily on Social Security to support them in
retirement.
I. Data and Methods
We rely on 30 years of data (19802009) from
the March CPS in our analysis of labor market conditions on labor force status and Social
Security receipt. The CPS is the leading survey
of labor market activity in the United States.
The monthly CPS survey asks questions about
the respondents involvement in the labor market around the time of the survey and collects
demographic data. In March of each year, a supplement is administered that includes questions
about income receipt, including from Social
Security, in the preceding calendar year. Each
March CPS provides sample sizes of 130,000 to
215,000. We focus on those between the ages of
55 and 69. As we describe subsequently, to be
Discussants: Colleen Flaherty Manchester, University
of Minnesota; Scott Weisbenner, University of Illinois; Jon
M. Bakija, Williams College; Purvi Sevak, Hunter College.
* Coile: Department of Economics, Wellesley College,
Wellesley, MA 02481 (e-mail: [e-mail]);
Levine: Department of Economics, Wellesley College,
Wellesley, MA 02481 (e-mail: [e-mail]).
23
24
AEA PAPERS AND PROCEEDINGS
consistent with our analysis of the Census/ACS
data, we also restrict the sample to men. This
leaves us with a final sample size of 292,093
men ages 55 to 69 over the period 1980 through
2009.
Our analysis of income for older individuals,
focusing on those in their 70s, relies on the 2000
Census and the 2001, 2002, and 20062009
ACS.1 The Census provides a very large number of observations: 5 percent of the US population. To obtain time series variation, we augment
these data with the ACS data. The ACS is modeled after the Census, with similar variables and
coding, and contains data for 1 percent of the
population (starting in 2005; earlier samples are
smaller). We use unified Census/ACS extracts
from the Minnesota Population Center IPUMS
USA project (Steven Ruggles et al. 2009).
We place three sample restrictions on our data.
First, our focus is on retirement income, so we
restrict the sample to those individuals who have
already left the labor force. This is not a major
constraint given the age composition of the
sample. More importantly, we restrict our attention to men. We do so because women in these
birth cohorts are likely to receive Social Security
payments based on their husbands earning
records, either because they do not qualify for
benefits on their own or because their dependent
spouse benefits are greater. It will thus be the
market conditions present around the time that
the husband retired that matter, but for those
women who have become widowed, we do not
know the age of the husband. Finally, we focus
on those who report some income from Social
Security. With over 90 percent of respondents in
their 70s in this category, those who report no
such income are likely not eligible to receive it,
so their income from this source would not be
altered by labor market conditions. We are left
with a final sample of 570,679 men age 70 to 79.
We augment the CPS and Census/ACS samples
with state-level unemployment rate data from
the Bureau of Labor Statistics.
MAY 2011
Our statistical models are identified based on
the differing historical experiences of individuals born into different birth cohorts. In essence,
we treat labor market conditions around the time
of retirement as a draw that is randomly assigned
to individuals. In our analysis of labor force status and Social Security receipt using March CPS
data, the availability of state-level data on labor
market conditions enables us to implement traditional quasi-experimental methods in regression models that include both state and year
fixed effects along with age fixed effects to control for differential patterns in outcomes by age.
We estimate linear probability models where the
outcome variables indicate whether an individual is out of the labor force and whether he is
collecting Social Security benefits as a function
of the state unemployment rate in the preceding
calendar year (to allow for the change in labor
force status to have occurred by the survey date),
holding constant these fixed effects along with
demographic characteristics (race/ethnicity,
marital status, educational attainment). Because
the availability of Social Security benefits at
age 62 suggests that responses to a weak labor
market may differ by age, we interact the unemployment rate with a series of categorical age
variables. It is our hypothesis that individuals
will be more likely to withdraw from the labor
force in response to a recession if they are at
least 62 years old. Moreover, any impact on
Social Security receipt should not begin until
that age since retirement benefits are not available before then.2
Our models of income for retirees in their
70s similarly rely on the plausibly exogenous
variation generated by whether an individual is
unlucky enough to approach retirement age during a recession. We estimate regression models
where the dependent variable is the natural log
of Social Security income, and the key explanatory variables are the average historical unemployment rates at earlier ages (5557, 5859,
6061, 6264, and 6569). The unemployment
rates that an individual experienced can almost
be thought of as a historical accident. Along
1
J. Trend Alexander, Michael Davern, and Betsey
Stevenson (2010) caution users about the potential that age
and sex variables in the 2000 Census and some subsequent
ACSs may include some miscoded data due to erroneous
disclosure avoidance procedures. To date, released versions
of the 2000 Census and 2006 ACS contain corrected age
data. We do not use data from the 2003 through 2005 ACS
surveys because no corrected data are available.
2
The impact may begin somewhat before age 62 because
the survey does not distinguish whether Social Security
receipt is attributable to retirement or disability. David H.
Autor and Mark G. Duggan (2003) have shown that some
older individuals rely on Social Security disability benefits
to provide income support when times are tough.
RECESSIONS, REtIREMENt, AND SOCIAL SECuRItY
VOL. 101 NO. 3
25
Table 1Impact of Labor Market Conditions on Older Workers Labor Force Status and Social Security Receipt
(Standard errors in parentheses)
Attended/completed
college
All
Unemployment rate age 6061
Unemployment rate age 6264
Unemployment rate age 6569
Panel B. Outcome: Social Security receipt
Unemployment rate age 5859
Unemployment rate age 6061
Unemployment rate age 6264
Unemployment rate age 6569
Observations
High school
graduates
0.083
(0.138)
0.294
(0.160)
1.091
(0.184)
1.221
(0.186)
0.214
(0.296)
0.641
(0.294)
1.113
(0.318)
1.433
(0.297)
0.238
(0.236)
0.350
(0.264)
1.323
(0.233)
1.770
(0.218)
0.116
(0.180)
0.085
(0.229)
0.514
(0.290)
0.344
(0.294)
0.040
(0.089)
0.181
(0.147)
0.894
(0.205)
0.220
(0.167)
0.177
(0.266)
0.508
(0.290)
1.218
(0.305)
1.070
(0.232)
0.039
(0.148)
0.063
(0.188)
0.741
(0.221)
0.503
(0.232)
0.104
(0.091)
0.053
(0.144)
0.175
(0.287)
0.739
(0.242)
292,093
Panel A. Outcome: Out of labor force
Unemployment rate age 5859
High school
dropouts
77,387
95,350
119,356
Notes: All estimates come from linear probability models that also include the current and lagged annual unemployment rate
without age interactions, race/ethnicity, marital status, and exact age, state of residence, and year fixed effects. Reported coefficients are multiplied by 100. Standard errors are clustered at the state level.
with these unemployment rates, we also control
for the contemporaneous unemployment rate
and state, age, and survey year fixed effects. The
idea is to see if labor market conditions around
the time of retirement (and we are agnostic
regarding exactly what that age is) affect the
income of retirees ten or so years later.
II. Results
Table 1 presents the results of our analysis
of March CPS data examining impact the of
labor market conditions on labor force status
and Social Security receipt. The top panel of the
table focuses on the likelihood that men between
the ages of 55 and 69 have dropped out of the
labor force. We examine this outcome for all
men in this age group, as well as for subsamples
by level of educational attainment (high school
dropouts, high school graduates, and those who
have attended any college regardless of whether
they have graduated). Each model interacts last
years annual average unemployment rate with
a set of age categories to assess whether labor
market conditions have a differential impact
on labor force status at different ages. The
base group for this comparison is men age 55
to 57, so these interactions can be interpreted
as the differential impact of unemployment on
the specified age group relative to this younger
group. Since the younger group is less likely to
have retired, it serves as something like a control
group that can capture the broader link between
aggregate labor market conditions and an individuals labor force status, abstracting from the
impact on retirement.
The results indicate that a higher unemployment rate leads to a greater probability of
withdrawal from the labor force (retirement)
as workers age, particularly after age 62 when
Social Security benefits are available. A one percentage point increase in the unemployment rate
is estimated to increase the number of 62- to 64and 65- to 69-year-old workers who are out of
the labor force by over a full percentage point.
The effects are stronger for high school dropouts
and high school graduates than for those with at
least some college education. The p-value on a
26
MAY 2011
AEA PAPERS AND PROCEEDINGS
Table 2Long-Term Impact of Labor Market Conditions on the Amount of Social Security Received by Retirees
(Standard errors in parentheses)
Panel A. Outcome: Social Security income
Unemployment rate at age 5557
Unemployment rate at age 5859
Unemployment rate at age 6061
Unemployment rate at age 6264
Unemployment rate at age 6569
Panel B. Outcome: total income
Unemployment rate at age 5557
Unemployment rate at age 5859
Unemployment rate at age 6061
Unemployment rate at age 6264
Unemployment rate at age 6569
Observations
High school
graduates
Attended/completed
college
All
High school
dropouts
0.186
(0.069)
0.003
(0.105)
0.024
(0.102)
0.435
(0.121)
0.591
(0.143)
0.095
(0.114)
0.203
(0.139)
0.275
(0.140)
0.517
(0.177)
0.194
(0.193)
0.153
(0.117)
0.056
(0.127)
0.062
(0.143)
0.396
(0.116)
0.557
(0.162)
0.206
(0.179)
0.099
(0.220)
0.179
(0.255)
0.206
(0.300)
1.115
(0.383)
0.018
(0.096)
0.084
(0.084)
0.156
(0.131)
0.109
(0.135)
0.248
(0.273)
0.131
(0.184)
0.122
(0.172)
0.372
(0.202)
0.332
(0.217)
0.379
(0.327)
0.001
(0.129)
0.169
(0.138)
0.087
(0.174)
0.025
(0.205)
0.346
(0.344)
0.306
(0.140)
0.009
(0.162)
0.212
(0.217)
0.133
(0.226)
0.083
(0.256)
570,679
154,710
205,848
210,121
Notes: Estimates come from linear regressions where the dependent variable is the natural log of Social Security benefits
received. Other control variables include the contemporaneous annual unemployment rate, race/ethnicity, marital status, and
age, state of residence, and year fixed effects. Reported coefficients are multiplied by 100. Standard errors are clustered by
state.
test of the equality of coefficients at ages 62 to
64 across groups is 5.7 percent.
We find similar patterns when we focus our
attention, in the lower panel of the table, on the
likelihood of Social Security receipt. Again,
a higher unemployment rate increases Social
Security receipt beginning at age 62 and mainly
for workers in the two lower educational attainment groups. Interestingly, the magnitude of the
effect at ages 62 to 64 is larger than it is at ages
65 to 69 for Social Security receipt, but that is
not true for labor force status. A plausible explanation is that as workers age into their late 60s,
they commence receipt of Social Security benefits regardless of their labor force status, lessening the importance of labor market fluctuations
on receipt at those older ages.
Table 2 reports the results of our analysis of
the effect of labor market conditions around the
time of retirement on the income of retired men
ages 70 to 79. We focus on income received
from Social Security in the top panel and on
total personal income in the bottom panel. In
both cases, we report coefficients on the historical unemployment rates that these individuals
experienced in their late 50s through their 60s,
using the average rate at specified ages. Models
are estimated for all men and for subsamples by
level of educational attainment.
The results suggest that experiencing a recession in the years leading up to retirement lowers
subsequent retirement income. This finding is
particularly evident for Social Security income,
for less educated workers, and for labor market
events that take place at or after age 62. Overall,
experiencing a one percentage point increase in
the unemployment rate at ages 62 to 64 reduces
income from Social Security by 0.435 percent.
VOL. 101 NO. 3
RECESSIONS, REtIREMENt, AND SOCIAL SECuRItY
We interpret this magnitude subsequently. As
anticipated, the effect of unemployment at ages
62 to 64 on retiree income falls as education
rises, although imprecision prevents us from
rejecting the null hypothesis that the coefficients are equal. One puzzling finding in this
table is that higher unemployment at ages 65 to
69 is associated with significantly higher Social
Security benefits for retired men in their 70s.
Although we have no specific explanation for
this finding, we are at least somewhat comforted
by the fact that the coefficient for high school
dropouts is small and statistically insignificant.
When we focus on total income, we find that
the only significant effect of historical unemployment on subsequent retiree income is the
effect of unemployment at ages 60 to 61 for
high school dropouts. That coefficient has a
p-value of 0.085. The equivalent coefficient for
age 6264 unemployment is of a comparable
magnitude, but it is not quite significant at even
the 10 percent level. Overall, the findings from
Table 2 provide some support for our hypotheses, but they are not as strong as they could be.
A final way to interpret these findings is by
assessing whether the magnitude of the coefficients is reasonable. To address this, we can
combine results from the bottom panel of Table 1
regarding Social Security receipt around retirement age and the top panel of Table 2 regarding
Social Security income for workers in retirement. Consider, for instance, the impact of a one
percentage point increase in the unemployment
rate at ages 62 to 64 for high school dropouts.
The results in Table 1 suggest that they would be
1.218 percentage points more likely to receive
Social Security at that age. The results in Table 2
suggest that subsequent Social Security income
in retirement is estimated to fall by 0.517 percentage points. If that one-point increase in
unemployment led 1.218 percent more workers to collect Social Security at ages 62 to 64
and that change led to a 0.517 percent reduction in aggregate receipt of Social Security
income, this suggests that the average worker
who began Social Security receipt as a result
of the increase in unemployment experienced a
42 percent reduction (0.517/1.218 = 0.424) in
Social Security benefits. A similar analysis for
high school graduates indicates that the average incremental Social Security recipient would
face a 53 percent benefit reduction. To put these
estimates in perspective, for individuals in these
27
birth cohorts, claiming benefits at 62 rather than
65 would be associated with a 20 percent reduction in their Social Security benefit.3 On that
basis, our estimates are rather high. However,
we have estimated 95 percent confidence intervals for both estimates using the delta method,
and both include this 20 percent figure.
III. Conclusions
This analysis began by hypothesizing that the
current high rate of unemployment in the United
States will lead some older individuals to withdraw from the labor force early (particularly
after age 62), to claim Social Security benefits
early, and to subsequently receive lower retirement income (as a result of earlier claiming of
Social Security benefits). Our analysis of historical data provides evidence supporting much
of this conjecture. We find that individuals age
62 and older are more likely to withdraw from
the labor force when the unemployment rate is
high, and are also more likely to be receiving
Social Security benefits. These effects are strongest for less educated workers. Our analysis of
retiree income is less persuasive, but still offers
evidence that is suggestive in its support of our
hypotheses. Taken as a whole, we believe that
these findings highlight the importance of focusing on older workers who have lost their jobs
in the present recession and may suffer lower
income for many years as a result.
REFEREnCES
Alexander, J. Trent, Michael Davern, and Betsey Stevenson. 2010. Inaccurate Age and Sex
Data in the Census PUMS Files: Evidence and
Implications. national Bureau of Economic
Research Working Paper 15703.
Autor, David H., and Mark G. Duggan. 2003. The
Rise in the Disability Rolls and the Decline in
Unemployment. Quarterly Journal of Economics, 118(1): 157205.
3
These rules have since changed so that now a worker
who retires at 62 rather than 66 receives a 25 percent reduction in their benefit. In fact, the effect of being laid off and
claiming three years earlier than expected could be bigger
than 20 percent if the worker also lost the opportunity for
several high-earnings years to be incorporated into the calculation of his Social Security benefit.
28
AEA PAPERS AND PROCEEDINGS
Coile, Courtney C., and Phillip B. Levine. 2007.
Labor Market Shocks and Retirement: Do
Government Programs Matter? Journal of
Public Economics, 91(10): 190219.
Coile, Courtney, and Phillip B. Levine. 2009. The
Market Crash and Mass Layoffs: How the Current Economic Crisis May Affect Retirement.
national Bureau of Economic Research Working Paper 15395.
MAY 2011
Coile, Courtney C., and Phillip B. Levine. 2010.
Reconsidering Retirement: How Losses and
Layoffs Affect Older Workers. Washington,
DC: Brookings Institution Press.
Ruggles, Steven, Matthew Sobek, Trent Alexander,
Catherine A. Fitch, Ronald Goeken, Patricia Kelly
Hall, Miriam King, and Chad Ronnander. 2009.
Integrated Use Microdata Series: Version 4.0.
Minneapolis: Minnesota Population Center.
Copyright of American Economic Review is the property of American Economic Association and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express
written permission. However, users may print, download, or email articles for individual use.