Early Retirement and the Housing “Bubble”:
Preliminary Evidence
May 2006
Martin Farnham
University of Victoria
Purvi Sevak*
Hunter College
Abstract
We use data from the Health and Retirement Study (HRS) and the Office of Housing Enterprise Oversight
to measure the effect of changes in housing wealth on retirement timing. Using cross-MSA variation in
house-price movements to identify wealth effects on retirement timing, we find evidence that such wealth
effects are present. We find that the rate of transition into retirement increases in the presence of positive
housing wealth shocks. According to our estimates, a 10% increase in housing wealth is associated with an
approximate 6% increase in the annual probability of retirement. In addition, we use data on expected age
of retirement from the HRS to measure the impact of housing wealth shocks on expectations about
retirement timing. Using these data, we find that a 10% increase in housing wealth is associated with
retiring approximately 2 months earlier. Some evidence suggests an asymmetric response to housing wealth
changes, with gains causing early retirement, but losses having little to no impact on retirement timing.
*Corresponding author: psevak@hunter.cuny.edu
I. Introduction
House prices in the US rose by an average of 40% from 1995-2004 (Himmelberg,
Mayer, and Sinai, 2005). This run-up in housing wealth produced many millionaires,
increased availability of secured credit to many households, and has led to recent
speculation that a housing “bubble” may be about to burst. Economists disagree on
whether current pricing in US housing markets constitutes a bubble—loosely defined as a
condition where asset prices exceed levels suggested by economic fundamentals. Case
and Shiller (2003) present evidence suggestive of a bubble in housing markets. To the
contrary, Himmelberg, Mayer, and Sinai (2005) argue that in spite of the significant runup in house prices, the user cost of housing remains within its historical range in most
real estate markets, and therefore there is little evidence of a bubble. However they do
note that user cost of housing is highly sensitive to increases in interest rates, so that a
substantial, lasting increase in interest rates could lead to large housing price declines.
Regardless of whether one believes that housing is overpriced in the US, the last
decade has demonstrated that significant changes in housing wealth can occur over a
relatively short period of time. History suggests that housing prices (in real terms) can
also decline for significant periods of time. And whether one believes prices are set to fall
due to the bursting of a bubble or due to rising interest rates, the possible effects of a
decline in house prices on household and individual behavior are worth considering.
Housing wealth effects are of interest to labor economists more generally because
housing constitutes a large portion of the assets of a typical American family, and is
therefore likely to be an important source of wealth effects on labor supply. Such wealth
effects are also of direct interest to policymakers. In the US and other western countries,
there has been a long-run trend towards early retirement. This trend, coupled with an
increase in life expectancy, has put public pension systems under strain, due to reduced
payroll tax collections resulting from early retirement.
Since housing assets represent an important share of the typical US household’s
total wealth, the labor supply behavior of older individuals may be closely linked to
changes in housing wealth. Understanding the empirical relationship between housing
wealth and labor supply is of interest to economists and is important to policymakers
seeking to better understand the financial well being of older households, and those
1
seeking to forecast the fiscal balance of public pension programs. It may also be of
interest to researchers studying the elderly home equity-consumption puzzle, in which
older households are found to consume (goods and services) very little out of housing
wealth in retirement (Venti and Wise, 2000).
Changes in housing wealth should be expected to affect consumption in a way
similar to other changes in permanent income. Standard lifecycle consumption theory
predicts that permanent, unanticipated shocks to wealth should result in adjustment of
consumption of goods and services as well leisure. Assuming well-functioning capital
markets, greater wealth should increase an individual’s consumption of normal goods and
services as well as their consumption of leisure, which we assume to be normal.
Individuals may access increased housing wealth through financial products such as
home equity loans or reverse mortgages. They may also access this wealth through
informal borrowing markets. For instance, an elderly couple may intend to bequeath their
home to their children when they die. If the couple experiences capital gains on their
home but finds it difficult to access this newfound wealth through formal financial
markets, they may turn to their children for financial assistance, on the understanding that
the size of the children’s expected inheritance has risen. Or households that were
targeting certain bequest levels may be able to increase consumption of goods, services,
and leisure if bequest targets are met sooner than expected due to housing wealth gains.
Therefore even without perfectly functioning formal financial markets, individuals may
be able to access newfound housing wealth through informal capital markets that
facilitate increased consumption.
Of course, if such formal and informal capital markets are missing, then housing
wealth changes may have little to no impact on consumption behavior. Given the
existence of capital market imperfections, the presence of housing wealth effects is an
empirical question. A number of studies have tested for housing wealth effects on the
consumption of goods and services. To date, little if any work has been done on housing
wealth effects on the consumption of leisure. In this paper we test for housing wealth
effects on the timing of retirement.
We use data from the 1992-2002 waves of the University of Michigan’s Health
and Retirement Study (HRS). This biennial survey provides detailed data on
2
demographic, financial, and labor supply characteristics of individuals who were between
age 51 and 61 in 1992. Therefore the panel allows for study of a population with
substantial housing wealth that is nearing retirement age.
Using restricted access geocodes for the HRS, we merge house-price indices at
the level of Metropolitan Statistical Area (MSA) from the Office of Federal Housing
Enterprise Oversight (OFHEO) to the HRS data. This allows us to generate a proxy for
changes in housing wealth that varies across MSAs. We use these data rather than HRS
data on housing wealth, due to the high degree of measurement error in the HRS housing
wealth data.
In the findings we present here, we find evidence of economically significant
housing wealth effects on retirement timing. We find that workers with greater
percentage gains in housing wealth since 1992 tend to retire sooner than other workers.
We also find that among non-retired workers, the expected age of retirement declines in
the face of housing wealth gains. Finally, we observe an apparently asymmetric response
of retirement timing to gains versus losses in housing wealth. While workers facing
housing wealth gains retire earlier, workers facing losses do not appear to significantly
alter their retirement timing.
II. Review of Existing Wealth Effects Literature
Economists have long been interested in the effects of wealth on household and
individual consumption behavior. A number of studies have sought to ascertain the
impact of stock market wealth on consumption of goods and services (see Poterba, 2000
for a survey). In the past two decades, several authors have tried to measure the effect of
housing wealth on consumption of goods and services. Some of these studies have been
done at the macro level (e.g. Bhatia 1987; Case, et al. 2005; Otto and Voss 2005). Others
have used micro data to measure housing wealth effects (e.g. Skinner 1996; Engelhardt
1996; Disney, et al. 2003; Belsky and Prakken 2004; Lehnart 2004; Bostic, et al. 2005).
The sharp rise in real estate markets in recent years has raised the profile of this question
in the empirical wealth effects literature.
Wealth effects on labor supply decisions are difficult to identify because wealth is
not randomly assigned. Some of the variation in wealth reflects individual heterogeneity
3
in preferences that will also be correlated with labor supply decisions. In the context of
retirement behavior, high levels of wealth could be a result of plans to retire early. All
else equal, an individual who plans to retire sooner than another individual should save
more during her working years, because she will have more years in retirement during
which she must live off of her accumulated wealth. Alternatively, wealthy individuals
may have strong tastes for work and thus retire later. Thus, cross-sectional estimates of
the effect of wealth on retirement timing could be biased in either direction.
The early literature suggests such a bias. In his survey of men’s labor supply,
Pencavel (1986) writes, “Of the 57 different estimated coefficients on net worth…only 16
would be judged as significantly different from zero…of these 16, exactly one-half is
positive and one-half is negative…this hardly constitutes a resounding corroboration of
the conventional static model of labor supply.” Although voluminous, the literature has
failed to find consistent evidence of wealth effects. This may partly be due to the
measures of income or wealth used in these papers. Studies often used measures of
unearned income, such as spousal earnings, dividends, rent, and government transfer
payments, to estimate the effect of wealth on labor supply (Kosters, 1966; Ashenfelter
and Heckman, 1974), but these income sources are likely to be endogenous to the
individuals’ labor supply decisions.
A number of more recent papers have examined wealth effects due to arguably
exogenous policy changes, but continue to provide mixed results. Hurd and Boskin
(1984) find that the Social Security benefit increases from 1969 to 1972 can explain a
large amount of the acceleration of retirement in that period, whereas Burtless (1986),
using the same data, finds that the effects were very small. Kruger and Pishke (1992)
estimate the effect of reductions in benefits due to amendments to Social Security in 1977
and find no effect. A more recent study by Chan and Stevens (2004) finds that
individuals’ planned retirement ages do respond to perceived changes in pensions.
Several studies avoid the problem of endogeneity by making use of natural
experiments. Imbens, Rubin, and Sacerdote (2001) survey lottery winners and find
significant labor supply effects of winnings, particularly among individuals ages 55 to 65.
A paper by Kimball and Shapiro (2001) making use of survey questions on hypothetical
lottery winnings finds large responses to wealth gains. Holtz-Eakin, Joulfaian and Rosen
4
(1993) find that individuals in households that receive large inheritances are more likely
to leave the labor force and less likely to enter the labor force.
Though a number of papers have exploited the stock market “bubble” of the late
1990s to test for wealth effects on retirement, results are inconclusive. Sevak (2005) and
Coronado and Perozek (2001) find earlier retirements among individuals who were most
likely to gain from the performance of the stock market – stock holders or those with
defined contribution pension plans, but Hurd and Reti (2001) find that the stock market
had no significant effect on retirement.
There is strong evidence that an individual’s retirement timing responds to
expected changes in wealth. Accrual rates in Social Security (Coile and Gruber, 2000)
and private defined benefit pension plans (Stock and Wise, 1990; Samwick, 1998) have
been found to be important determinants of retirement timing. Although these effects are
not referred to as wealth effects, they provide evidence of the sensitivity of retirement
timing to the path of expected wealth accumulation.
A related question is whether housing wealth affects the consumption of leisure.
Compared to the empirical literature on wealth effects on consumption of goods and
services, relatively little empirical work has been done to measure housing wealth effects
on leisure. To our knowledge only one paper has directly studied the impact of housing
wealth on retirement timing. Doling and Horsewood (2003) use aggregate data from a
cross-section of European countries to argue that the declines in retirement age can be
attributed to increased homeownership rates. However, they do not attempt to directly
measure a housing wealth effect.
The housing boom of the 1990s provides an opportunity to reexamine the
question of wealth effects on retirement timing. To the extent that the boom in real estate
markets can be considered an exogenous shock to wealth (we discuss this further below),
we can use variation in housing price changes across regions to identify a housing wealth
effect on retirement timing.
In addition, the housing boom itself presents potential challenges to policymakers.
To the extent that individuals respond to increases in housing wealth by dropping out of
the labor force, pressure on public pension systems due to declining payroll tax receipts
may increase. To the extent that decisions to exit the labor force are not fully reversible,
5
then any future retrenchment in the housing market may have important impacts on the
financial security of older households. To the extent that housing wealth effects on
retirement timing have been large during the housing boom, the risk associated with
retrenchment may also be large. Therefore, beyond providing a potential natural
experiment from which to learn about wealth effects, the housing boom of the 1990s in
itself represents a phenomenon worthy of study.
III. Data
In order to examine the effect of changes in housing wealth on the retirement
timing of older individuals, we assemble a data set with observations on individual
workers from the University of Michigan’s Health and Retirement Study (HRS) for the
years 1992-2002 and match them to data on MSA-level house prices from the Office of
Federal Housing Enterprise Oversight (OFHEO) and county-level unemployment rate
data from the Bureau of Labour Statistics (BLS). The HRS contains extensive data on
income, wealth, work and health status, and other individual and household
characteristics of individuals near retirement age. In addition, the survey codes
geographic identifiers in each wave that allow matching to the local-level data from
OFHEO and BLS.
An overview of the HRS data is given at
http://hrsonline.isr.umich.edu/data/index.html. Because the HRS is an ongoing panel
survey, construction of an analysis file can be a complex process. A baseline set of 7,650
randomly selected households with a member born between 1951 and 1961 were
interviewed in 1992. Every two years, the HRS attempts to re-interview the members of
this household. Though the re-interview rate is quite high (see Table 1, discussed below),
in some cases a household or individual household member may refuse to conduct an
interview or may be unreachable. If this is the case, the HRS continues to attempt to
interview them in subsequent waves. The HRS provides a “tracker” file, which
summarizes the interview status of each household member over time and facilitates the
merge of multiple years of data. It includes data on whether the person was interviewed,
and reasons for non-interview including whether the individual has died or is in a nursing
home.
6
In each interview year, respondents answer detailed questions about current and
past labor supply, health, and other topics. A designated “financial respondent” in each
household provides detailed financial information on the household. This includes values
of various assets, including housing, real estate, stocks, bonds, checking and savings
accounts, individual retirement accounts (IRAs), small businesses, and pensions. Since
housing assets are reported at the household level and cannot be attributed to one
particular spouse, we measure housing wealth and all other household wealth at the
household level and assume that households pool their resources.
While we use an individual’s 1992 level of housing equity as a control in the
specifications discussed below, we do not use measures of changes in housing wealth
derived from HRS data. This is because the HRS housing wealth data are plagued with
measurement error. Much of this measurement error is due to coding errors, and hence
does not reflect an individual’s perception of his own housing wealth.1 As would be
expected, estimation of housing wealth effects on retirement timing using these errorladen housing wealth data from the HRS typically yields coefficients of zero.2
Since year of retirement is available on an annual basis, and yet HRS surveys are
conducted biennially, we annualize the dataset by imputing values of other individual
characteristics for the odd-numbered years (in which respondents are not surveyed). The
imputation of values is done linearly using data from the previous and next survey years.
For example, values for 1995 would be imputed linearly from 1994 and 1996 survey
values. Timing of retirement does not need to be imputed, because respondents report the
year they retired.
The stock of retirees among HRS respondents, expressed as a percentage of the
analysis sample, is plotted in Figure 1. At age 52 (the youngest age of HRS respondents
in our analysis sample) approximately 18% of the sample self-identifies as retired. By age
61 this has risen to around 46%. By age 71 (the oldest age of HRS respondents in our
1
While individuals have been shown to misperceive their housing wealth to a substantial
degree, we might not wish to discount such errors of perception, given that an individual
bases his decisions on his or her (mis)perceptions.
2
As an example of how poor the housing wealth data in the HRS are, we found that from
1992-2002, the median cumulative increase in housing wealth was just 1.06%. Clearly
this deviates dramatically from the overall housing market experience of the typical
American household over that period of time.
7
sample) this percentage has reached nearly 100%. Transitions into retirement by age,
expressed as percentage of the analysis sample, are shown in Figure 2, where we see that
transitions into retirement accelerate as individuals enter their fifties. There is a spike in
retirement transitions around age 62, presumably corresponding to the start of eligibility
for Social Security benefits, and a subsequent spike at 65. Transitions into retirement then
taper off after age 65.
Table 1 illustrates our sample selection. We focus on men because men of this
generation tend to have had a stronger attachment to the labor force than women. Since
retirement may be an ill-defined concept for someone who works mostly in the home, we
focus our analysis on men, for whom we expect retirement to be better defined. In the
first survey wave (1992) 4,605 men born between 1931-1941 were selected to be
surveyed. These men show up in the sample every year thereafter, though their data may
be missing due to attrition for various reasons. The sample remaining in the workforce
(non-retired) declines over time, as would be expected. Of these non-retired individuals,
those who remain in the sample and report working also decline over time. Further
sample reductions occur when we limit workers to those who are not self-employed,
those who are homeowners, those who have not moved to date, and those for whom the
HRS has valid county codes (and hence can be matched to local unemployment data).
While our sample selection may appear quite severe, once we limit the sample to
men, most further sample selection involves removal of individuals to whom our
empirical question does not apply. Questions of housing wealth effects on retirement
timing clearly do not apply to non-homeowners or non-workers. These groups account
for the vast majority of the observations we omit from our analysis. We also omit selfemployed individuals because we expect their behavior with respect to retirement timing
to differ substantially from that of wage and salary workers. We currently omit movers
from our sample. Future work will address movers, as they may be particularly
responsive to housing wealth shocks in cases where they have the option to move to areas
with a lower cost of living. We may also extend the analysis to include women.
Summary statistics for our analysis sample are given in Table 2. The analysis
sample includes male workers who are not self-employed, who are from a home-owning
household, and who have not yet moved since 1992. The mean age of respondents in our
8
sample is 59 years. The mean level of home equity (in 2002 dollars) in 1992 is $77,000,
among HRS respondents in our sample. According to OFHEO data matched to HRS
individuals, the average cumulative percent change in house value in the sample is
14.4%. The mean unemployment rate in counties where HRS respondents live is 6% over
the sample period.
Table 3 gives a more detailed distribution of the household wealth levels and
changes. The median level of home equity in 1992 was $56,000. The median change in
house value since 1992, was 8.5%. Note that figures in Table 3 encompass observations
between 1992 and 2002. Considering only changes in house value between 1992 and
2002, the mean and median changes are 54% and 50%, respectively (not shown in Table
3).
IV. Empirical Strategy
Measuring housing wealth effects on retirement timing is subject to a number of
challenges. First, housing wealth may be correlated with local labor market conditions.
Local home values may reflect current or future labor market opportunities, which may
also be correlated with retirement timing. If older individuals delay retirement in the face
of high wages and if high wages are capitalized into local house values, then failure to
control for local labor market conditions may lead to downward bias in estimates of
wealth effects. As a result, in the specifications reported below we control for the
worker’s wage and for the local (county-level) unemployment rate.
Second, housing wealth, defined as the value of equity in one’s home, is likely to
be endogenous. Households, to the extent they can liquidate equity in their home through
borrowing, may be able to adjust their housing wealth in ways correlated with their
preference for leisure. Households anticipating early retirement may engage in less
liquidation of home equity for other consumption purposes than households anticipating
later retirement. To the extent that home equity or changes in home equity reflect active
management of one’s housing wealth, these measures may reflect expected retirement
timing of households. This would potentially lead to upward bias in measurement of
housing wealth effects. As a result, rather than measuring the effect of changes in home
equity on retirement timing, we measure the effect of changes in house value on
9
retirement timing. Since, among non-movers, these are arguably exogenous changes in
household wealth, estimation of housing wealth effects based on change in house value
should not be subject to this upward bias.
Third, self-reported data on house values, while providing variation in wealth at
the individual level, is subject to classical measurement error. This will lead to downward
bias in the estimates of wealth effects on retirement timing. Initial work with the HRS
self-reported house value data suggests that measurement error is severe. An extensive
discussion of mismeasurement in self-reported data on house values can be found in
Engelhardt (2005). As a result of the measurement error obviously present in the HRS,
we have chosen to use OFHEO house price indices at the MSA level, which are matched
to HRS data using restricted-access geocodes.
Fourth, the permanent income hypothesis suggests that the response of
consumption to changes in wealth should be greatest for permanent, unexpected changes
in wealth. In a standard model of lifetime consumption, the consumption path (whether of
goods and services or leisure) is predetermined and therefore unresponsive to expected
wealth shocks. Furthermore, to the extent that individuals are lifetime consumption
smoothers, changes in wealth that are only transitory should have a smaller effect on
lifetime consumption than changes in wealth that are permanent. There is substantial
long-term variation in the rate of house price appreciation across metro areas in the US
(Gyourko, Mayer, and Sinai, 2004). This suggests that expectations of capital gains on
housing should vary across metro areas. Properly isolating unexpected and permanent
changes in housing wealth is important for correctly measuring wealth effects on
retirement timing. In the current draft, we do not address this particular challenge, though
we discuss this point in the context of anticipated further work in our concluding
remarks.
The first basic specification we use in this preliminary analysis is a linear
probability model of entry into retirement. We use this model primarily for the sake of
ease of interpretation, and because it allows us to difference out unobserved
heterogeneity by inclusion of fixed effects. For this basic specification, we use the
following model of entry into retirement:
10
(1)
'
DitRETIRE = Wi,1992
! + "#HPI i,t $1992 + Xit' % + uit .
DRETIRE is a dummy variable equal to 1 if individual i retires at time t and equal to zero
otherwise. W is a vector of household wealth variables. Measures of household wealth
include housing wealth, business wealth, and other wealth in 1992. Inclusion of these
variables is intended to form a combined measure of permanent income at the start of the
sample period. ΔHPI, the variable of primary interest, denotes change in housing wealth.
It is measured as the cumulative percent change in housing wealth since 1992, calculated
from the OFHEO MSA-level house price indices.
Changes in housing wealth are measured in percentage terms, rather than in dollar
terms. We choose this specification of the wealth variable because we expect that relative
changes in housing wealth matter more than absolute changes. In other words, a $50,000
increase in housing wealth will probably have a larger impact on the retirement timing
choice of someone with a $100,000 home in 1992 (for whom it is a 50% increase in
housing wealth) than someone with a $1,000,000 home in 1992 (for whom it is only a 5%
increase). To be precise, a 5% cumulative change in housing wealth would be denoted as
“5” in the data.
X is a vector of control variables, including individual-level characteristics,
household characteristics, and measures of local labor market conditions. Controls
included in our specifications include age, years of education, race, lagged marital status,
a lagged indicator of poor health, and lagged log wage. We also include year fixed effects
and controls for lagged industry and lagged occupation. In some specifications we add
the county-level unemployment rate and state fixed effects.
Estimates of this model are performed on the analysis sample described in Section
III, using OLS. Standard errors are clustered at the individual level to account for the fact
that we have multiple observations on individuals. Results from this specification are
given in Table 4.
Another way to consider the effect of changes in housing wealth on retirement
timing is to exploit data in the HRS on expected age of retirement. Workers in each
survey wave of the HRS are asked at what age they intend to retire. Since worker
expectations vary at a greater frequency than actual retirement status (transition to
11
retirement can only occur given the way we structure our analysis sample), we may be
able to measure housing wealth effects on retirement timing with greater precision using
these data. In addition, the use of retirement expectations allows us to observe responses
to housing wealth effect before actual retirement occurs. This eliminates the problem of
right censoring that occurs with individuals in our sample whose future retirement
(beyond the sample period) may, in fact, be affected by housing wealth changes, but
whose retirement we never observe. This right censoring could be expected to lead to
downward bias in our estimates of wealth effects on actual retirement transitions. The
annual within-individual variation expected retirement age also allows for inclusion of
individual-level fixed effects, which may reduce bias due to unobserved individual
heterogeneity.
To estimate the impact of housing wealth changes on expected retirement
timing—namely expected retirement age—we estimate variants of the following
empirical model of expected retirement age:
(2)
'
E[Ageitretire ] = Wi,1992
! + "#HPI i,t $1992 + Xit' % + vit
Right-hand-side variables are the same as in Equation (1). The left-hand-side
variable is the age, in years, at which the respondent (at time t) expects to retire.
Estimates of this model are performed on the analysis sample described in Section
III, using OLS. Standard errors are clustered at the individual level to account for the fact
that we have multiple observations on individuals. Results from this specification are
given in Table 5. Results of a specification that includes individual fixed effects are given
in Table 6.
V. Estimation Results
Table 4 gives estimates of equation (1), the model of transition into retirement.
The estimated coefficient on the cumulative house-price-change variable in Column 1 is
0.0006 and is statistically significant at the five-percent level. The magnitude of the
coefficient estimates suggests that a 10% increase in housing wealth leads to an increase
of 0.006 in the annual retirement probability. Given a base retirement probability of 0.108
12
in our analysis sample, this amounts to a 5.5% increase in the annual retirement
probability for a 10% gain in house value.
Addition of the county-level unemployment rate has little impact on the estimated
wealth effect, as can be seen in Table 4, Column 2. Addition of state fixed effects,
increases the magnitude of the coefficient on housing wealth change so that it implies that
a 10% increase in house prices results in a 6.5% increase in the probability of retirement.
This estimate is statistically significant at the ten-percent level. Overall, results in Table 4
are consistent with the presence of a housing wealth effect on retirement timing.
It is interesting to note that there appears to be an asymmetric response by
individual workers to housing gains versus housing losses. Appendix Table 1 uses the
same specifications as in Table 4, but splits the measure of cumulative percent change in
house prices since 1992 into a measure cumulative percent increase and a measure of
cumulative percent decrease. Studies in the housing mobility literature (e.g. Engelhardt,
2005) have found that households respond asymmetrically to housing wealth gains and
losses. Typically, studies find that declines in house prices lead to reduced mobility,
while gains have no effect on mobility. This can be for reasons of nominal loss aversion
or equity constraints. Since residential mobility and retirement tend to be somewhat
correlated, we were interested to see whether housing losses lead to reduced probability
of retirement, while gains had no effect.3
As Appendix Table 1 demonstrates, we find the opposite effect. Gains increase
the annual probability of retirement on the order of 7.5%, while losses have no
statistically significant effect on retirement timing. This finding deserves further
investigation, but on the surface suggests that workers have more flexibility to retire early
than they do to retire late.
Turning now to the question of wealth effects on expected age of retirement,
Table 5 gives estimates of Equation 2. Equation 2 models expected retirement age as a
function of changes in housing wealth. Note that the sample size shrinks to 6478. This is
due primarily to the fact that while our analysis sample includes observations of
3
Mobility and retirement tend to be correlated because many households postpone moves
(especially long-distance moves) until retirement. Moving to a retirement destination
prior to retirement would require an impractical job search in many cases.
13
households immediately following retirement, such households obviously do not report
an expected age of retirement.
Results in Column 1 suggest that a 10% cumulative increase in house prices since
1992 leads to a decline in expected retirement age of 0.2 of a year (2.4 months). Inclusion
of the county unemployment rate and state fixed effects do not substantially change our
findings, as can be seen in Table 5, Columns 2-3. In all cases, the estimated housing
wealth effects are statistically significant at the ten-percent level.
Results of re-estimation of Equation 2 with individual-level fixed effects are
given in Table 6. Here we see that while the magnitude of the wealth effect coefficient
declines somewhat, it remains statistically significant. These coefficient estimates imply
that a 10% increase in housing wealth leads to approximately a 2 month decline in
expected retirement age. Again, inclusion of county-level unemployment rate has little
effect on our estimates.
As in the case of actual retirement transitions, it appears that housing gains and
losses have an asymmetric impact on expected retirement age. Appendix Table 2 uses the
same specifications as in Table 5, but splits the measure of cumulative percent change in
house prices since 1992 into a measure cumulative percent increase and a measure of
cumulative percent decrease. Estimates suggest that while individuals respond to housing
wealth increases by reducing their expected retirement age, they do not increase their
expected retirement age in response to housing wealth decreases.
VI. Conclusion and Further Work
In this preliminary work we find evidence of modest housing wealth effects on
retirement timing. We find that the annual probability of retirement rises by 5.5-6.5% for
a 10% cumulative increase in house prices. Gains in house price appear to have a greater
effect on transitions into retirement than losses.
We also find evidence that expectations about retirement timing are affected by
cumulative changes in housing wealth. Our estimates suggest that a 10% increase in
house prices leads individual workers to reduce their expected retirement age by 2-2.5
months. While extrapolating to bigger housing gains requires caution, our findings raise
14
the possibility that individuals in areas with large house-price gains may have
substantially revised their retirement timing as a result of housing wealth shocks.
Our findings suggest that the run-up of housing prices over the last decade may
have had important impacts on labor supply in a number of housing markets. It also
suggests that a sudden decline in house prices could reverse this process to some extent,
as households reverse their plans for early retirement.
There are a number of potential problems with this preliminary analysis. We omit
movers from our analysis sample. Movers may have a greater ability to access housing
wealth gains. Those making local moves may be able to trade down, thus liquidating
some of their gains. Those making longer distance moves may be able to liquidate even
more of their gains, if they move to areas with lower housing costs. Thus, further analysis
will consider the behavior of movers, and treat the retirement and residential mobility
choices as joint decisions. We have not yet developed a way to properly measure
unexpected gains in housing wealth. Since many gains may be anticipated, failure to
separate expected from unexpected gains may reduce measured wealth effects.
In further work, we also plan to address the issue of expected versus unexpected
housing wealth gains. Since expected gains in housing wealth should not affect
consumption of leisure (in the absence of borrowing constraints), estimates of wealth
effects are likely to be attenuated if we fail to correctly measure unexpected gains. Fairlie
and Krashinsky (2006) use Current Population Survey (CPS) Outgoing Rotation Group
(ORG) files to generate MSA-level measures of house price appreciation. Following
Hurst and Lusardi (2004) they regress house price changes at the MSA level on changes
in state and local economic indicators and demographic variables and include four years
prior appreciation in an MSA’s housing prices to further capture anticipated housing
gains. They use residuals from this regression as proxies for unanticipated house price
appreciation at the MSA level. A similar approach could prove useful as we proceed with
this work.
Finally, we plan to explore the mechanism by which house-price gains lead to
earlier retirement. For instance, are households accessing new housing wealth through
financial markets, as in the case of home equity borrowing? Or, if households are saving
for bequests, do housing gains allow them to reach bequest targets sooner and hence
15
allow their members retire earlier? Or can individuals in households with larger housing
gains retire earlier because they realize they can rely on children (whose expected
bequests have risen) to serve as substitutes for precautionary saving? We hope to provide
some answers to these questions in further work on this project.
16
References
Ashenfelter, Orley C. and James J. Heckman (1974). “The Estimation of Income and
Substitution Effects in a Model of Family Labor Supply.” Journal of Labor
Economics, 17(4), p757-783.
Bostic, Raphael, Stuart Gabriel and Gary Painter (2005). “Housing Wealth, Financial
Wealth, and Consumption: New Evidence from Micro Data.” manuscript, University
of Southern California.
Burtless, Gary (2000). “How Would Financial Risk Affect Retirement Income Under
Individual Accounts?” An Issue in Brief; Center for Retirement Research at Boston
College (5).
Case, Karl E. and Robert J. Shiller (2003). “Is There a Bubble in the Housing Market?”
Brookings Papers on Economic Activity, v2003, n2.
Chan, Sewin and Ann Huff Stevens (2004), “Do Changes in Pension Incentives Affect
Retirement? A Longitudinal Study of Subjective Retirement Expectations.” Journal
of Public Economics, v88(7-8): p1307-1333.
Coile, Courtney C. and Jonathan Gruber (2000). “Social Security and Retirement” NBER
Working Paper #7830.
Coronado, Julia Lynn and Maria Perozek, "Wealth Effects and the Consumption of
Leisure: Retirement Decisions During the Stock Market Boom of the 1990s,"
Unpublished Manuscript, 2001.
Disney, Richard, Andrew Henley and David Jevons (2003). “House price shocks,
negative equity and household consumption in the UK in the 1990s,” manuscript,
University of Wales Aberystwyth.
Doling, John and Nick Horsewood (2003). “Home ownership and early retirement:
European experience in the 1990s.” Journal of Housing and the Built Environment.
Vol. 18, p289-308.
Engelhardt, Gary V. (2005). “Nominal Loss Aversion, Housing Equity Constraints, and
Household Mobility: Evidence from the United States.” Journal of Urban Economics
(forthcoming).
Engelhardt, Gary V. (1996). “House prices and home owner saving behavior.” Regional
Science and Urban Economics, 26, p313-336.
Fairlie, Robert W. and Harry A. Krashinsky (2006). “Liquidity Constraints, Household
Wealth, and Entrepreneurship Revisited.” manuscript, UC Santa Cruz.
17
Gyourko, Joseph, Christopher Mayer, and Todd Sinai (2004). “Superstar Cities.”
Columbia Business School and Wharton School working paper.
Himmelberg, Charles, Christopher Mayer, and Todd Sinai (2005). “Assessing High
House Prices: Bubbles, Fundamentals and Misperceptions.” NBER Working Paper
#11643.
Holtz-Eakin, Douglas, David Joulfaian, and Harvey S. Rosen (1993). “The Carnegie
Conjecture: Some Empirical Evidence,” The Quarterly Journal of Economics, 108
(2), p413-435.
Hurd, Michael D. and Michael J. Boskin (1984). “The Effect of Social Security on
Retirement in the Early 1970s.” The Quarterly Journal of Economics 99(4), p767790.
Hurd, Michael D. and Monika Reti, “The Effects of Large Capital Gains on Work and
Consumption: Evidence from Four Waves of the HRS” RAND Report for the U.S.
Department of Labor, October 2001.
Imbens, Guido W., Donald B. Rubin, and Bruce Sacerdote (2001). “Estimating the Effect
of Unearned Income on Labor Supply, Earnings, Savings, and Consumption:
Evidence from a Survey of Lottery Winners.” The American Economic Review 91(4),
p778-795.
Kezdi, Gabor and Purvi Sevak (2004). “Economic Adjustment of Recent Retirees to
Adverse Wealth Shocks.” Michigan Retirement Research Center Working Paper.
Kimball, Miles and Matthew Shapiro (2003). “Labor Supply: Are the Income and
Substitution Effects Both Large or Both Small?” University of Michigan Unpublished
Manuscript.
Kosters, M. (1966). “Income and Substitution Effects in a Family Labor Supply Model.”
Rand Corporation Report #P-3339.
Krueger, Alan B. and Jorn-Steffen Pischke (1992). “The Effect of Social Security on
Labor Supply: A Cohort Analysis of the Notch Generation,” Journal of Labor
Economics 10(4), p412-437.
Pencavel, John. “Labor Supply of Men: A Survey.” In Handbook of Labor Economics,
Vol. 1. Edited by Orley C. Ashenfelter and Richard Layard. North Holland: Elsevier
Science B.V. 1986.
Poterba, James M. (2000). “Stock Market Wealth and Consumption.” The Journal of
Economics Perspectives, Vol. 14, No. 2, p99-118.
18
Samwick, Andrew A. (1998). “New Evidence on Pensions, Social Security, and the
Timing of Retirement.” Journal of Public Economics 70(2), p207-236.
Sevak, Purvi (2005). “Wealth Shocks and Retirement Timing: Evidence from the
Nineties.” Hunter College Working Paper.
Skinner, J. S. (1996). “Is housing wealth a sideshow?” Advances in the Economics of
Aging, National Bureau of Economic Research Report, p241-268. University of
Chicago Press.
Stock, James H. and David A. Wise (1990). “Pensions, the Option Value of Work, and
Retirement.” Econometrica 58(5), p1151-1180.
Venti, Steven F. and David A. Wise (2000). “Aging and Housing Equity.” National
Bureau of Economic Research Working Paper #7882.
19
Figure 1.
Percent HRS Respondents In Analysis Sample Who Have Transitioned to Retirement, 1992-2002
100%
90%
80%
70%
% Retired
60%
50%
40%
30%
20%
10%
0%
52
54
56
58
60
62
Age
20
64
66
68
70
Figure 2.
Percent of HRS Respondents in Analysis Sample, by Age at Which they First Retire
25%
20%
% Retiring
15%
10%
5%
0%
52
54
56
58
60
62
Age
21
64
66
68
70
Table 1. Sample Selection Among Male HRS Respondents Born Between 1931 and 1941
Year
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
Eligible
Men
4,605
4,605
4,605
4,605
4,605
4,605
4,605
4,605
4,605
4,605
4,605
Total
50,655
Not Retired at Working at
Not SelfHomeowner
t-1
t-1
Employed at t-1
at t-1
Has Not Moved
1992-2002
Has No
Missing Data
3,357
3,154
2,911
2,691
2,445
2,231
2,023
1,813
1,655
1,495
3,120
2,942
2,377
2,179
1,835
1,647
1,368
1,187
960
825
2,494
2,343
1,883
1,709
1,444
1,290
1,065
909
731
614
1,965
1,843
1,475
1,328
1,171
1,038
868
741
595
499
1,965
1,717
1,358
1,154
1,012
859
700
578
454
353
1,934
1,668
1,294
1,102
963
781
672
524
430
348
23,775
18,440
14,482
11,523
10,150
9,716
22
Table 2. Summary Statistics for Analysis Sample
Age
Years of Education
Hispanic
Black
Married (t-1)
Poor Health (t-1)
Log Wage (t-1)
Retiree Health Ins (t-1)
Expected Retirement Age
Home Equity in 1992
Business Equity in 1992
Other Wealth in 1992
Cumulative House Price Index Change Since 1992
County Unemployment Rate
Mean
58.90
12.64
0.07
0.12
0.90
0.12
2.75
0.62
63.46
76,811
13,892
92,589
14.37
0.06
Sample Size
9,716
23
Standard Deviation
3.48
3.19
0.26
0.33
0.29
0.33
0.58
0.49
4.90
77,218
145,843
249,467
19.37
0.03
Table 3. Detailed Distribution of Selected Financial Variables
Home Equity in 1992
Business Equity in 1992
Other Wealth in 1992
Cumulative % Change in House Price Index Since 1992
Mean
76,811
13,892
92,589
14.37
Sample Size
9,716
24
25th Percentile
30,500
0
13,000
2.17
Median
56,000
0
37,600
8.54
75th Percentile
100,000
0
91,750
23.03
Table 4.
OLS Estimates: Effect of Changes in House Prices on Transition to Retirement
Years of Education
Hispanic
Black
Married (t-1)
(1)
-0.001
(2)
-0.001
(3)
-0.001
(0.001)
(0.001)
(0.001)
-0.006
-0.005
-0.006
(0.013)
(0.013)
(0.015)
-0.014
-0.014
-0.013
(0.01)
(0.01)
(0.01)
-0.036
**
(0.012)
Poor Health (t-1)
0.027
Has Retiree Health Ins. (t-1)
**
%Δ House Prices (t-1992)
0.009
0.009
(0.008)
0.085
-0.042
**
N
R-squared
**
*
0.084
*
(0.051)
**
-0.042
0.026
**
(0.012)
0.078
-0.041
0.024
(0.014)
0.00063
**
0.00070
(0.0003)
(0.0004)
-0.040
-0.055
(0.113)
(0.134)









0.108
9,716
0.077
0.108
9,716
0.077
0.108
9,716
0.081
** Denotes statistical significance at the 5% level and * at the 10% level. Robust standard errors are in parentheses.
25
**
(0.014)
0.022
**
**
(0.055)
(0.015)
0.00064
**
(0.006)
0.022
County Unemployment Rate
% Retiring
0.027
(0.006)
(0.015)
(0.0003)
Age and Year Dummies
Industry and Occupation Codes
County Unemployment Rate
State Fixed Effects
0.029
(0.007)
0.027
**
(0.011)
0.009
(0.012)
Other Wealth (1992, $1000s)
**
(0.007)
(0.051)
Business Equity (1992, $1000s)
0.027
-0.037
(0.011)
(0.011)
(0.006)
Home Equity (1992, $1000s)
**
(0.012)
(0.011)
Log Wage (t-1)
-0.036
*
*
Table 5.
OLS Estimates: Effect of Changes in House Prices on Expected Retirement Age (in years)
Years of Education
Hispanic
Black
Married (t-1)
Poor Health (t-1)
Log Wage (t-1)
(1)
0.114
**
Home Equity (1992, $1000s)
Business Equity (1992, $1000s)
Other Wealth (1992, $1000s)
%Δ House Prices (t-1992)
**
(3)
0.103
(0.037)
(0.037)
(0.037)
-0.084
-0.193
-0.739
(0.379)
(0.392)
(0.459)
-0.374
-0.385
-0.333
(0.234)
(0.235)
(0.248)
0.385
0.390
0.341
(0.291)
(0.29)
(0.3)
-0.201
-0.196
-0.179
(0.215)
(0.215)
(0.215)
-0.588
**
(0.188)
Has Retiree Health Ins. (t-1)
(2)
0.115
-0.715
-0.585
**
(0.187)
**
-0.716
-0.636
**
-0.706
(0.159)
(0.156)
-1.619
-1.609
-1.658
(1.315)
(1.313)
(1.336)
-0.077
-0.076
-0.073
(0.421)
(0.418)
(0.454)
0.048
0.056
-0.022
(0.256)
(0.257)
(0.253)
(0.006)
County Unemployment Rate
**
-0.020
**
-0.017
(0.006)
(0.007)
3.127
-0.522
(3.478)
(4.383)
Age and Year Dummies
Industry and Occupation Codes
County Unemployment Rate
State Fixed Effects









Mean Expected Retirement Age
63.5
6,478
0.203
63.5
6,478
0.204
63.5
6,478
0.215
N
R-squared
** Denotes statistical significance at the 5% level and * at the 10% level. Robust standard errors are in parentheses.
26
**
(0.188)
(0.159)
-0.020
**
**
**
Table 6. Household Fixed Effects Estimates
of Effect of Changes in House Prices on Expected Retirement Age (in years)
Married (t-1)
Poor Health (t-1)
Log Wage (t-1)
Has Retiree Health Ins. (t-1)
%Δ House Prices (t-1992)
-0.005
-0.006
(0.472)
(0.472)
-0.259
-0.259
(0.24)
(0.24)
-0.094
-0.094
(0.223)
(0.223)
-0.062
-0.062
(0.144)
(0.144)
-0.015
(0.007)
County Unemployment Rate
**
-0.015
**
(0.007)
-0.64866
(6.385)
Mean Expected Retirement Age
N
R-square
63.5
6,483
0.039
63.5
6,483
0.038
** Denotes statistical significance at the 5% level and * at the 10% level. Robust standard errors are in parentheses.
27
Appendix Table 1.
OLS Estimates: Effect of Increases and Decreases in House Prices on Retirement Transition
Years of Education
Hispanic
Black
Married (t-1)
(1)
-0.001
(2)
-0.001
(3)
-0.001
(0.001)
(0.001)
(0.001)
-0.008
-0.006
-0.006
(0.013)
(0.014)
(0.015)
-0.013
-0.012
-0.013
(0.01)
(0.01)
(0.01)
-0.036
**
(0.012)
Poor Health (t-1)
Log Wage (t-1)
Has Retiree Health Ins. (t-1)
0.029
**
% Increase in House Prices (t-1992)
% Decrease in House Prices (t-1992)
0.030
0.009
0.009
0.009
(0.007)
(0.007)
(0.008)
0.086
-0.043
**
0.027
**
(0.006)
*
0.086
*
(0.051)
**
-0.042
0.026
**
(0.012)
0.077
-0.040
0.023
0.024
(0.014)
0.00083
**
0.00081
(0.0003)
(0.0003)
(0.0004)
0.00050
0.00055
-0.00007
(0.001)
(0.001)
(0.001)
-0.056
-0.061
(0.114)
(0.134)









9,669
0.078
9,669
0.078
9,669
0.082
% Retiring
N
R-squared
** Denotes statistical significance at the 5% level and * at the 10% level. Robust standard errors are in parentheses.
28
**
(0.014)
(0.015)
**
**
(0.055)
0.023
0.00084
**
(0.006)
(0.015)
County Unemployment Rate
Age and Year Dummies
Industry and Occupation Codes
County Unemployment Rate
State Fixed Effects
**
(0.011)
0.027
**
(0.011)
(0.011)
(0.012)
Other Wealth (1992, $1000s)
0.029
-0.038
(0.011)
(0.051)
Business Equity (1992, $1000s)
**
(0.012)
(0.006)
Home Equity (1992, $1000s)
-0.037
*
*
Appendix Table 2.
OLS Estimates: Effect of Increases and Decreases in House Prices on Expected Retirement Age
Years of Education
Hispanic
Black
Married (t-1)
Poor Health (t-1)
Log Wage (t-1)
(1)
0.117
**
Home Equity (1992, $1000s)
Business Equity (1992, $1000s)
Other Wealth (1992, $1000s)
% Increase in House Prices (t-1992)
% Decrease in House Prices (t-1992)
**
(3)
0.106
(0.037)
(0.037)
(0.037)
-0.073
-0.193
-0.753
(0.378)
(0.392)
(0.458)
-0.396
*
-0.410
*
(0.238)
(0.249)
0.396
0.403
0.356
(0.29)
(0.29)
(0.3)
-0.196
-0.190
-0.171
(0.215)
(0.215)
(0.215)
-0.586
**
-0.715
-0.582
**
(0.187)
**
-0.716
-0.634
**
-0.712
(0.159)
(0.155)
-1.644
-1.635
-1.644
(1.322)
(1.32)
(1.334)
-0.089
-0.089
-0.096
(0.42)
(0.417)
(0.451)
0.045
0.054
-0.035
(0.256)
(0.257)
(0.254)
**
-0.023
**
-0.023
(0.008)
(0.008)
(0.008)
0.002
-0.001
-0.019
(0.016)
(0.016)
(0.016)
3.487
-0.125
(3.515)
(4.416)
County Unemployment Rate
Age and Year Dummies
Industry and Occupation Codes
County Unemployment Rate
State Fixed Effects









Mean Expected Retirement Age
63.5
6,477
0.204
63.5
6,477
0.204
63.5
6,477
0.215
N
R-squared
** Denotes statistical significance at the 5% level and * at the 10% level. Robust standard errors are in parentheses.
29
**
(0.187)
(0.159)
-0.023
**
-0.353
(0.237)
(0.188)
Has Retiree Health Ins. (t-1)
(2)
0.117
**
**