123
Cityscape: A Journal of Policy Development and Research • Volume 9, Number 3 • 2007
U.S. Department of Housing and Urban Development • Ofce of Policy Development and Research
Cityscape
Homeowner Age and
House Price Appreciation
David T. Rodda
Freddie Mac
Satyendra Patrabansh
Abt Associates Inc.
Abstract
Do the houses of elderly homeowners appreciate at the same rate as the average house
in their local market? As the population ages and retirees plan their financial future,
homeowners need to project accurately the value of their single largest asset—their
house. The federal government is also concerned about the financial welfare of its
elderly citizens and the solvency of the insurance for reverse mortgages. Using Health
and Retirement Study data, we find that the houses of elderly (75 years old or older)
homeowners appreciate 1 percentage point less per year in real terms than the houses
of middle-aged (50 to 74 years old) homeowners. These estimates are smaller than
the findings of Davidoff (2004), who used the American Housing Survey to show a
3-percentage-point slower house appreciation rate for homeowners aged 75 or older
relative to that of all other homeowners. Using census microdata in nonlongitudinal
form (1990 and 2000), we find 2.4-percentage-point slower real house appreciation for
elderly homeowners. Houses of elderly homeowners thus appreciate in real terms at a
1- to 3-percentage-point discount relative to their local markets.
Introduction
Do the houses of elderly homeowners appreciate at the same rate as the average house in their local
market? The answer matters most directly to elderly homeowners making long-range financial
plans. For most elderly homeowners, and especially for low-wealth homeowners, their house is
Funding for this research was provided by the U.S. Department of Housing and Urban
Development (HUD), Office of Policy Development and Research, Contract C-OPC-
21452 RG. Additional funding was provided by Abt Associates Inc. through a Daniel T.
McGillis Development and Dissemination Grant. The opinions expressed in this paper are
the authors’ own and do not represent HUD, Abt Associates, or Freddie Mac.
124
Refereed Papers
Rodda and Patrabansh
their largest asset. It would be logical to assume that the house would appreciate at the long-run
average house price appreciation rate (5.9 percent per year in nominal terms minus 4.1 percent
for inflation equals 1.8 percent in real terms).
1
Despite this assumption, elderly homeowners
might have lower house appreciation rates because they spend less money on remodeling and
maintenance. Most people know of an old person who has lived in the same house for many years
and done little to update the property. Does this anecdotal evidence represent an outlier or should
elderly homeowners expect lower house price appreciation?
Elderly homeowners are not the only ones concerned about their house values and financial
planning. Certainly their children have a vested interest in providing for their parents. Local
governments rely on property taxes linked to house values. Towns with a high proportion of
elderly homeowners have to provide sufficient social services, particularly for seniors who prefer
to stay in their own homes. Both the families and their local governments want to preserve the
older properties as a source of affordable housing for the next generation of homebuyers. Finally,
the federal government cares about elderly homeowners and their house values. In particular, the
Federal Housing Administration insures Home Equity Conversion Mortgages (HECMs), which
enable elderly homeowners to convert their house equity into cash. The homeowners do not need
to pay off the reverse mortgage until they move or permanently leave their home; at that point,
the house is sold to pay off the loan. The long-run viability of the HECM insurance fund depends
on projected house values exceeding loan balances. Given the potentially long time horizon of
20 years or more before the loan is paid, what should government officials assume about future
house price appreciation?
Using data from Growing Older in America: The Health and Retirement Study (HRS), we find that
the houses of elderly (75 years or older) homeowners appreciate in constant dollars at a rate 1.0 to
1.2 percentage points less per year than the houses of middle-aged (50 to 74 years old) homeown-
ers. This discount in house price appreciation for age is smaller than the 3-percentage-point dis-
count estimated in Davidoff (2004) using American Housing Survey (AHS) data. Our HRS estimate
compares house value appreciation for elderly homeowners with that of middle-aged homeowners
for a period of 10 years or less, while Davidoff’s (2004) AHS estimates contrast house value ap-
preciation for elderly homeowners with that for all homeowners and follows the homeownership
period for up to 16 years. After adjusting the HRS estimates for the 0.4-percentage-point gap
between middle-aged homeowners and all homeowners, our best estimate for the elderly discount
in house value appreciation is 1.4 to 1.6 percentage points relative to all homeowners.
Both the HRS and AHS data sets are longitudinal. Longitudinal data sets enable us to control for
fixed personal and property-specific effects and provide the best way to track the change in house
values without the confounding effects of the changing composition of households and properties.
The appreciation rate can also be tracked using the Census Bureau’s Public Use Microdata Sample
(PUMS). PUMS has a disadvantage in that it consists of two independent cross-sections, 1990
and 2000, which were aggregated by age group and metropolitan statistical area (MSA). PUMS
data have large samples and better controls for building age and length of tenure than those that
1
The 5.9-percent annual house price appreciation rate comes from the Office of Federal Housing Enterprise Oversight
(OFHEO) national House Price Index from the first quarter of 1975 to the first quarter of 2005. The inflation rate of 4.1
percent is calculated from the Consumer Price Index excluding shelter expenses for the same period.
125
Cityscape
Homeowner Age and House Price Appreciation
are available in the HRS data. PUMS data indicate that the house prices of homeowners 75 years
old and older appreciate 2.4 percentage points more slowly per year than do those of younger
homeowners.
Thus, two independent studies, this one and Davidoff’s (2004), analyzing three separate data
sets—HRS, AHS, and PUMS—show a negative and significant relationship between age and house
value appreciation. The elderly discount from the HRS is about half the discount from the AHS or
PUMS, but even that smaller difference could be critical in the long run to homeowners, lenders,
and insurance funds (see exhibit 5 in the second section following this introduction).
The remainder of the article is divided into four sections. The first section provides a brief review
of the literature on elderly housing decisions. The second section presents the HRS data used
for analysis and ends with estimations of house value appreciation using HRS/Assets and Health
Dynamics Among the Oldest Old (AHEAD) Study data and comparisons with recent findings by
Davidoff (2004) using the AHS data. The third section provides a benchmark from the Census
Bureau’s PUMS data by comparing appreciation rates between 1990 and 2000. A final section summa-
rizes the findings in light of six alternative “stories,” each of which may contain some partial truth.
2
Literature Review
Several authors have addressed the issue of elderly wealth. Venti and Wise (2001a, 2001b) have
used HRS/AHEAD data to find that equity-rich, low-income households tend to reduce equity
when they sell. People with substantial amounts of nonhousing wealth shift their assets into
housing, whereas people with limited nonhousing wealth rebalance their portfolio by reducing
the housing equity share. Overall, housing equity increases until about age 75 and then declines
by about 1.76 percent per year. Homeowners with intact households rarely move or refinance to
take equity out of their house. The equity decline among older homeowners is driven primarily
by 7.84 percent of households experiencing a health shock (either a death or move to a nursing
home) to their family status.
The life-cycle model (Hurd, 1990) predicts that wealth will be “decumulated” as people age,
but the uncertainty about the amount of time until death leads people to spend down nonhous-
ing wealth first and hang onto their house as long as possible. In fact, the results confirm that
nonhousing wealth is spent down faster and earlier than home equity. Owned housing is not just
an investment; it also provides a stream of real consumption and precautionary savings against
unexpected costs, especially health costs (Heiss, Hurd, and Borsch-Supan, 2003).
Goodman and Thibodeau (1997, 1995) found that the largest reduction in house value from
depreciation occurs in the first 10 years of the building’s life before tapering off to 0 when the
building reaches age 20 and slightly increasing in years 20 to 40, presumably due to remodeling.
(See also Harding, Rosenthal, and Sirmans, 2007.) The findings by Goodman and Thibodeau
(1997, 1995) suggest that old buildings do not suffer greater depreciation as the homeowner
2
The full report, The Relationship Between Homeowner Age and House Price Appreciation, which includes policy implications
and statistical appendixes, is available on the Abt Associates Inc. website (http://www.abtassociates.com/reports/HP_Aging_
Final.pdf).
126
Refereed Papers
Rodda and Patrabansh
ages, but the extremely aged homeowners were probably a small share of the sample. Even more
consistent in their findings than the rate of depreciation is the widening variance in house values as
buildings age. The range of house values for an old building is usually much wider than for a new
house. Part of this age-related heteroskedasticity is due to home improvement projects, including
additions and remodeling. Capozza, Israelsen, and Thomson (2005) refer to the “atypicality” of
a house that has acquired unusual features as it has aged. Appraisers may have a difficult time
finding comparable houses in the neighborhood and therefore discount the appraised value. Older
homeowners often have not updated the style of their house for 10 to 20 years, and, as a result, the
house becomes atypical relative to other houses on the market. The dated styles lower demand and
increase the search time for a suitable buyer, leading to discounts for atypical houses. These find-
ings directly support the fifth explanation for elderly discounts: the higher variance of older house
values is a leading cause of lower appreciation rates for older homeowners.
The most relevant predecessor to this article is by Davidoff (2004). He used the panel of national
AHS data from 1985 to 2001 both to measure the house price appreciation of homes owned by the
elderly and to link maintenance spending to homeowner age. He found that elderly homeowners
spend less on maintenance. Homeowners who are more than 75 years old spend $270 less on routine
maintenance than do younger homeowners of similar homes and spend $1,100 less on all home
improvements. Older homeowners also realize lower house price appreciation by about 3 percent
per year than do younger homeowners for similar homes in similar markets during 1985 to 2001.
Most of the house value data that are publicly available use information reported by the homeown-
ers. How reliable are the self-reported house values by elderly homeowners? Using the national
AHS (1985 to 1987), Goodman and Ittner (1992) found that the average homeowner overvalues
his or her house by 6 percent, with an average absolute error rate of 14 percent. DiPasquale and
Somerville (1995) used the AHS data to compare the rate of appreciation in house prices based on
transaction units with the entire stock. They found that units with longer tenure had lower house
values and lower appreciation.
Kiel and Zabel (1999) used confidential metropolitan AHS data merged with census tract-level
information for the neighborhood around each unit. They found that recent buyers report house
values that are 8.4 percent higher than the eventual sales price, whereas homeowners with longer
tenure overvalue their houses by only 3.3 percent.
3
Kiel and Zabel (1999) estimated that the
self-reports were, on average, 5.1 percent too high, but the upward bias was not related to the
characteristics of the house, occupants (except for tenure), or neighborhood. Also, the upward
bias on homeowners’ valuations seems to decline with the length of tenure. When Kiel and Zabel
(1999) controlled for maintenance or remodeling, the difference between value and sales price fell
by 1 percentage point.
3
Fisher and Williams (2006) offer one explanation for high estimates of recent buyers. Basing their analysis on Consumer
Expenditure Survey data, they found a spike in additions and maintenance spending in the first 3 years of ownership.
127
Cityscape
Homeowner Age and House Price Appreciation
The Health and Retirement Study Data and Models
The HRS/AHEAD
4
is a particularly useful data source for investigating the relationship between a
homeowner’s age and the rate of house value appreciation. Many studies, particularly longitudinal
ones, have very little coverage of the elderly population because this segment is out of the labor
force and has previously represented a small share of the population. The HRS/AHEAD, however,
focuses on the near-retirement and elderly populations and surveys them roughly every 2 years.
The survey includes variables on family structure, living arrangements, retirement decisions,
financial state, and health status. As a source on housing data, it has not been used nearly as much
as the AHS. The HRS/AHEAD provides opportunities to corroborate findings from other studies
and further their analyses on elderly health and wealth issues.
The HRS tracks the same households as they enter retirement and experience the health challenges
of aging. HRS began with a longitudinal sample of more than 12,600 people in 7,600 households
who were born in the period from 1931 through 1941; the people in this sample were 51 to 61 years
old when the initial survey occurred in 1992. Followup surveys of the same households were
conducted in even years until 1998, when HRS was merged with AHEAD. AHEAD surveyed
7,447 people in about 6,000 households in which one member was born before 1923; that is,
one member of the household was 70 years old or older when the initial survey was conducted
in 1993. A followup survey was conducted in 1995. Both HRS and AHEAD oversampled African
Americans, Hispanics, and Florida residents.
5
In 1998, two new birth cohorts were added:
Children of the Depression Age (CODA), born between 1924 and 1930, and War Babies (WBs),
born between 1942 and 1947. Moreover, additions from new relationships and remarriages are
made in each followup survey. Therefore, the current HRS/AHEAD/CODA/WBs sample exceeds
22,000 people and 14,000 households; more than 18,000 people in more than 12,000 households
were interviewed in 2002.
Sample Selection
To select our sample, we used the HRS tracker and region files prepared in 2002 in conjunction
with the core HRS/AHEAD survey files of each even year from 1992 through 2002 and the years
1993 and 1995. We identified nearly 12,000 households that owned a single-family, nonfarm,
nonmobile,
6
noncondominium primary home in at least 1 survey year from 1992 through 2002.
House values were self-reported by the financial respondent in each wave. About 9,500 of those
households were observed in at least 2 survey years in the same primary home, providing two dif-
ferent snapshots of house values and other mutable characteristics of the same home and the same
homeowners at two distinct points in time. Because our unit of analysis is a primary home and
because some households are observed in more than one distinct primary home between 1992 and
4
The Health and Retirement Study (HRS) merged with the Assets and Health Dynamics Among the Oldest Old (AHEAD)
Study in 1998; now the studies are collectively referred to as HRS/AHEAD or simply HRS.
5
In part, because of this oversampling by race and location, weights are used throughout the analysis. If weights were
insufficient to restore the representativeness of the data, it is possible that the lower house price appreciation found in the
Health and Retirement Study results are linked to the sampling.
6
The Health and Retirement Study uses the term “mobile homes,” which we interpret as essentially synonymous with
manufactured housing. To be consistent with the survey coding, however, we refer to mobile homes in this article.
128
Refereed Papers
Rodda and Patrabansh
2002, the number of unique single-family, nonfarm, nonmobile, noncondominium owned primary
homes observed in at least 2 different survey years is 10,129.
7
After imputation for missing data
and the conversion of dollar amounts to 2002 dollars using the seasonally unadjusted Consumer
Price Index excluding shelter expenses for all urban consumers, we calculated the compound annual
growth rate (CAGR) of house values.
Description of HRS Data
Demographic characteristics are available from both the core survey files and the tracker file.
As recommended by HRS, we used the data from the tracker file as much as possible. To obtain
person information at the household level, we used the characteristics of the financial respondent.
8
When the financial respondent could not be identified or when the household did not participate
in the financial section of the surveys in the wave we were interested in, we used the characteristics
of the family respondent. We obtained fixed characteristics such as the date of birth, gender, race,
and ethnicity of the financial respondent or the family respondent from the tracker file. Some
demographic characteristics, such as whether respondents are in a nursing home and their coupled
status, change over time. In many instances, we used the characteristics from the end year, also
available from the tracker file. We also obtained the household weights from the tracker file.
In exhibit 1, we summarize demographic characteristics of homeowners. It should be noted that
some homeowners who are represented in multiple homes are counted multiple times in our sam-
ple of unique homes. No household appears more than three times and 88 percent of households
appear only once, as shown by the home sequence number in exhibit A-1 (see the appendix at the
end of the article). Almost all family respondents are financial respondents; that is, they answered
the financial sections of the survey.
We selected 75 years of age as our breakpoint for analysis because Venti and Wise (2001a, 2001b)
show that housing equity increases for homeowners until about age 75 and then declines. At both the
starting year and the ending year, only a few family respondents for the HRS/WBs group were 75 years
old or older and almost no family respondents for the AHEAD/CODA group were younger than 65.
The percentages of homeowners who were 75 years old or older in the start year are almost 0
for the HRS/WBs group and 36 for the AHEAD/CODA group; this comparison grows even more
pronounced in the end year. Even though some overlap occurs in the near-elderly age group of
65 to 74 years between the HRS/WBs and AHEAD/CODA groups, tabulating the characteristics of
the HRS/WBs group with those of the AHEAD/CODA group provides a good way to compare the
middle-aged or near-retirement family respondents with the elderly family respondents.
Besides homeowner age, the main distinctions in exhibit 1 between the middle-aged (HRS/WBs)
and the elderly (AHEAD/CODA) respondents are as follows:
7
We extracted values from the waves when the home was first and last observed and called them starting and ending house
values. Even though the Health and Retirement Study survey wave years are 1992, 1993, 1994, 1995, 1996, 1998, 2000,
and 2002, actual interview years range from 1991 to 2003. We call the actual interview years of first and last observation as
our start (interview) years and end (interview) years, respectively. Dollar adjustments to 2002 dollars are made on the basis
of interview years, not wave years. See the appendix for more details.
8
The financial respondent is the person responsible for overseeing the financial matters of the elderly person, often
the elderly person himself or herself and usually the same person as the family respondent. If the financial respondent
information was missing, the family respondent information (some other person from the same family) was substituted.
129
Cityscape
Homeowner Age and House Price Appreciation
Exhibit 1
Demographic
Characteristics
a
HRS/WBs
Respondents
AHEAD/CODA
Respondents
All
Respondents
Number Percent Number Percent Number Percent
Demographic Characteristics of Owners of Primary Homes (1 of 2)
Respondent type**
Financial 5,403 99.4 2,461 99.9 7,864 99.6
Family 36 0.6 3 0.1 39 0.4
Age in start year**
54 or younger 2,198 46.7 10 0.5 2,208 32.3
55–64 2,907 48.0 44 2.6 2,951 33.8
65–74 325 5.1 1,466 60.7 1,791 22.5
75–84 9 0.2 802 31.6 811 10.0
85 or older 0 0.0 142 4.6 142 1.5
Average age**
5,439 55.4 2,464 73.8 7,903 61.1
Median age
b
5,439 56.0 2,464 73.0 7,903 59.0
Age in end year**
54 or younger 506 11.5 3 0.1 509 7.9
55–64 2,968 59.3 16 1.1 2,984 41.2
65–74 1,847 27.1 477 20.4 2,324 25.0
75–84 116 2.0 1,487 61.0 1,603 20.4
85 or older 2 0.0 481 17.4 483 5.5
Average age**
5,439 61.0 2,464 79.1 7,903 66.7
Median age
b
5,439 62.0 2,464 79.0 7,903 66.0
• More HRS/WBs group households than AHEAD/CODA group households are couples
(71.5 compared with 41.2 percent).
• Fewer HRS/WBs group households than AHEAD/CODA group households are headed by
females (47.9 compared with 61.8 percent).
• Fewer HRS/WBs group households than AHEAD/CODA group households are headed by
Whites (89.0 compared with 93.4 percent).
The younger cohort is more likely to live in the South Atlantic Census Division (22.3 percent
compared with 18.3) and have much shorter average tenure (15.8 years compared with 26.0). In
terms of financial and health characteristics, the younger cohort is more likely to own a second
home (15.2 percent compared with 8.8) but has lower average liquid assets ($179,559 compared
with $192,241) and lower medical expenses ($2,055 compared with $2,837). HRS does not include
direct measurement of maintenance expenditures, but home improvements or major additions are
reported in exhibit 2. The younger cohort has a higher percentage of home improvements (26.9 percent
compared with 19.2) and a higher average biannual home improvement cost ($4,084 compared
with $2,826). Even though the older households have more liquid assets, they spend less on home
improvements and perhaps also on maintenance, which is not itself reported. House price appre-
ciation (measured by the CAGR in constant 2002 dollars) is shown in exhibit 2. The distributions
of growth rates range widely, but the average CAGR is significantly higher for the younger cohort
than for the elderly (2.28 percent compared with 1.52). Without a regression adjustment, the aver-
age house price appreciation is about 0.75 percentage point lower for the elderly homeowners.
130
Refereed Papers
Rodda and Patrabansh
AHEAD = Assets and Health Dynamics Among the Oldest Old Study. CODA = Children of the Depression Age. HRS = Health
and Retirement Study. WBs = War Babies.
*Indicates signicance at the 5-percent level.
**Indicates that the differences between the HRS/WB and AHEAD/CODA groups are statistically signicant at the 1-percent
level. χ
2
tests were conducted for cross-tabulation comparisons and t-tests were performed for average comparisons. The
observation level of the sample in this article is a primary home. Some households have as many as three primary homes in
the survey period from 1992 through 2002. Only single-family, nonfarm, nonmobile, and noncondominium owned primary
homes are considered in the sample, which is also conned to homes with nonimputed and nonmissing house values reported
by respondents in both start and end years. The sample sizes and medians presented in this exhibit are unweighted, but the
percentages and averages reported are weighted using household weights provided by HRS/AHEAD to make inference on the U.S.
population of the same age, gender, and race/ethnicity prole as that of the HRS/AHEAD sample. All dollar amounts are adjusted
to 2002 using the nonseasonally adjusted Consumer Price Index excluding shelter expenses.
a
From the tracker and region les, characteristics of the nancial respondent for each wave was obtained. Where there were
no nancial respondents, characteristics of the family respondent were obtained. Where the information on respondent type
was unavailable, characteristics of the oldest respondent were obtained. Information on homes, such as house values, were
obtained from the year-specic HRS/AHEAD survey data les.
b
Medians are calculated without weights and no statistical tests for signicance of the difference between the HRS/WBs and
AHEAD/CODA groups were conducted.
c
This variable captures whether the respondent or his or her spouse/partner was in a nursing home in the end year.
Source: 1992 to 2002 HRS/AHEAD
Years homeowner is age 75 or older
between start and end years**
0 5,357 98.5 665 28.1 6,022 76.5
1–5 74 1.4 1,151 46.4 1,225 15.4
6–10 8 0.1 648 25.5 656 8.1
Average years**
5,439 0.0 2,464 3.3 7,903 1
Median years
b
5,439 0.0 2,464 3.0 7,903 0.0
Coupled or partnered in end year** 3,843 71.5 1,028 41.2 4,871 62.0
In nursing home in end year**
c
18 0.3 74 2.8 92 1.1
Female** 2,772 47.9 1,509 61.8 4,281 52.2
Race**
White/Caucasian 4,504 89.0 2,226 93.4 6,730 90.4
African American 742 7.6 206 5.4 948 6.9
Other 176 3.2 27 1.0 203 2.5
Unknown 17 0.2 5 0.2 22 0.2
Ethnicity**
Mexican Hispanic 253 3.5 65 1.5 318 2.9
Other Hispanic 127 1.6 36 1.1 163 1.4
Non-Hispanic 5,042 94.8 2,359 97.3 7,401 95.5
Unknown 17 0.2 4 0.2 21 0.2
Exhibit 1
Demographic Characteristics of Owners of Primary Homes (2 of 2)
Demographic
Characteristics
a
HRS/WBs
Respondents
AHEAD/CODA
Respondents
All
Respondents
Number Percent Number Percent Number Percent
131
Cityscape
Homeowner Age and House Price Appreciation
AHEAD = Assets and Health Dynamics Among the Oldest Old Study. CAGR = compound annual growth rate.
CODA = Children of the Depression Age. HRS = Health and Retirement Study. WBs = War Babies.
* Values in this column are in percent unless designated by a dollar sign.
** Indicates that the differences between the HRS/WB and AHEAD/CODA groups are statistically signicant at the 1-percent
level. χ
2
tests were conducted for cross-tabulation comparisons and t-tests were performed for average comparisons. The
observation level of the sample in this article is a primary home. Some households have as many as three primary homes in
the survey period from 1992 through 2002. Only single-family, nonfarm, nonmobile, and noncondominium owned primary
homes are considered in the sample, which is also conned to homes with nonimputed and nonmissing house values
reported by respondents in both start and end years. The sample sizes and medians presented in this exhibit are unweighted,
but the percentages and averages reported are weighted using household weights provided by HRS/AHEAD to make
inference on the U.S. population of the same age, gender, and race/ethnicity prole as that of the HRS/AHEAD sample. All
dollar amounts are adjusted to 2002 using the nonseasonally adjusted Consumer Price Index excluding shelter expenses.
a
From the tracker and region les, characteristics of the nancial respondent for each wave were obtained. Where there were
no nancial respondents, characteristics of the family respondent were obtained. Where the information on respondent type
was unavailable, characteristics of the oldest respondent were obtained. Information on homes, such as house values, was
obtained from the year-specic HRS/AHEAD survey data les.
b
CAGR = (FV/PV)1/n – 1, where PV is the beginning value, FV is the ending value, and n is the number of intervening years.
This measure is very similar to ln(FV/PV)/n, which assumes continous compounding. We prefer CAGR because most house
price growth rates, like interest rate growth rates, are reported in annual growth rates.
c
Medians are calculated without weights and no statistical tests for signicance of the difference between the younger and
older groups were conducted.
Source: 1992 to 2002 HRS/AHEAD
Home Improvement or
Major Addition
a
Reported in end year**
Yes 1,388 26.9 463 19.2 1,851 24.5
No 4,039 72.9 1,995 80.6 6,034 75.3
Unknown 12 0.2 6 0.2 18 0.2
Average biannual home
improvement costs
5,392 $4,084 2,442 $2,826 7,834 $3,691
Median biannual home
improvement costs
5,392 $ 0 2,442 $ 0 7,834 $ 0
House Price Appreciation
CAGR
b
of primary home**
– 10.00% or less 173 2.8 116 4.6 289 3.3
– 10.01% to – 5.00% 271 4.8 135 5.5 406 5.0
– 5.01% to – 3.00% 222 3.9 130 4.9 352 4.2
– 3.01% to – 1.00% 848 14.8 484 19.2 1,332 16.2
– 1.01% to 0.00% 511 8.8 192 7.9 703 8.6
0.01% to 1.00% 507 8.8 219 9.1 726 8.9
1.01% to 3.00% 1,057 18.6 370 15.0 1,427 17.5
3.01% to 5.00% 680 12.9 275 11.5 955 12.5
5.01% to 10.00% 742 15.5 311 12.9 1,053 14.7
10.01% or more 428 9.1 232 9.5 660 9.2
Average CAGR of primary home**
5,439 2.28 2,464 1.52 7,903 2.04
Median CAGR of primary home
c
5,439 1.30 2,464 0.73 7,903 1.19
Exhibit 2
Home Improvement and House Price Appreciation
HRS/WBs
Respondents
AHEAD/CODA
Respondents
All
Respondents
Number Percent* Number Percent* Number Percent*
132
Refereed Papers
Rodda and Patrabansh
Estimation of House Value Appreciation Using HRS Data
In the previous section, we presented the HRS data and compared the HRS/WBs and AHEAD/CODA
cohorts, which have different age profiles. To isolate the age effect on home value appreciation and
control for other effects, such as demographic, tenure, geographic location, wealth, and cognition,
we perform ordinary least squares regression analyses. In our regressions, as in the tabulations, we
use weights provided by HRS to make inference on the U.S. population of the same gender and age
profile as the HRS sample. We use the survey regression commands in Stata
®
to correctly estimate
coefficients and standard errors for survey data because HRS contains survey data with 52 strata.
9
The first two columns of exhibit 3 replicate two models from Davidoff (2004) using the HRS
data. Because his final models control for building age and square footage, two variables that are
not available in HRS, we are able to replicate only two of his simpler models. In the first model,
Davidoff (2004) regresses the natural log of resale value divided by the 1985 value on the number
of years between 1985 and the resale year. Other control variables in the regression are an indicator
variable for the homeowner being 75 years old or older with interactions by MSA, and resale year
indicator variables. He shows that, for each additional year the homeowner is 75 years old or older,
the total appreciation of the house price between 1985 and the resale year decreases by 2.3 percentage
points. In his second model, Davidoff (2004) divides the log growth rate by the number of years
from 1985 to the year sold. The annualized growth rate is regressed on an indicator variable desig-
nating whether the homeowner was 75 years old or older in 1985 and the interactions of the MSA
and resale year indicator variables. This AHS regression shows that the homes of homeowners who
are 75 years old or older appreciate 2.2 percentage points less per year than the homes of younger
homeowners. We present Davidoff’s (2004) regression in exhibit 3 in columns (1) and (2).
Columns (3) and (4) present our replication of the AHS models using HRS data. Given that we did
not have resale value or year sold, we substituted self-reported house values in the start year and
the end year. In addition, instead of MSAs, we have census divisions as our location covariates.
10
The dependent variables are calculated as the natural logs of end value divided by start value and
we annualized the ratio by dividing by the number of elapsed years.
The coefficients for the HRS models are smaller than those for the AHS models. The HRS coefficients
show that, for each additional year the homeowner is 75 years old or older, the total appreciation
of the house value decreases by 0.7 percentage point and that homes of homeowners who are
75 years old or older appreciate 1.1 percentage points less per year than do homes of middle-aged
homeowners. Given that our sample has almost no homeowners younger than 45 in the start year
and 35 percent of homeowners in the AHS sample are younger than 45, our lower coefficients are
not surprising. The elderly homeowners will have smaller differences in house value appreciation
9
Unweighted regressions yielded nearly identical results.
10
The census division fixed effects control for differences in average appreciation rates by division, but they do not capture
the variation among metropolitan statistical areas (MSAs) within the division. If elderly homeowners disproportionately live
in MSAs with low house appreciation but the model omits controls for that MSA, it is possible that the lower appreciation
of the MSA will be transferred to the coefficient on the elderly homeowners. In fact, the Public Use Microdata Sample
models presented in the text and the exhibits show that the negative coefficient on elderly homeowners is even greater at
the MSA level.
133
Cityscape
Homeowner Age and House Price Appreciation
AHEAD = Assets and Health Dynamics Among the Oldest Old Study. AHS = American Housing Survey. CAGR = compound annual growth rate. HRS = Health and Retirement Study.
MSA = metropolitan statistical area.
**Indicates signicance at the 1-percent level.
a
AHS results are from Davidoff (2004).
b
The observation level of the sample in this report is a primary home. Some households have as many as three primary homes in the survey period from 1992 through 2002. Only single-
family, nonfarm, nonmobile, and noncondominium owned primary homes are considered in the sample, which is also conned to homes with nonimputed and nonmissing house values
reported by respondents in both start and end years. CAGR, the dependent variable, is a measure similar to annualized difference in natural logs of end and start values in (4). Results are
weighted to make inference on the U.S. population of the same age, gender, and race/ethnicity prole as that of the HRS/AHEAD sample using household weights provided by HRS. All
dollar amounts are adjusted to 2002 using the nonseasonally adjusted Consumer Price Index excluding shelter expenses.
c
Davidoff (2004) calls this variable YEARSa75. His start year is 1985 and end year is the actual year when the home was sold.
d
Davidoff (2004) calls this variable a75. His start year is 1985.
Source: 1992 to 2002 HRS/AHEAD
Years age 75 or older between
start and end years
c
– 0.023 – 0.0074 – 0.0023
(0.009) * (0.0018) ** (0.0003) **
Age 75 or older in start year
d
– 0.022 – 0.0107 – 0.0103
(0.016) (0.0029) ** (0.0028) **
Constant 0.167 0.026 0.239 0.043 0.047 0.046
(0.013) ** (0.004) ** (0.055) ** (0.013) ** (0.015) ** (0.015) **
Fixed effects MSA x
year sold
MSA x
year sold
Division x
end year
Division x
end year
Division x
end year
Division x
end year
N 2,781 2,757 7,309 7,309 7,309 7,309
R
2
0.36 0.30 0.08 0.07 0.06 0.06
Exhibit 3
AHS
a
HRS/AHEAD
b
(1) (2) (3) (4) (5) (6)
ln
Resale Value
In
Resale Value
ln
End Value
In
End Value
CAGR CAGR1985 Value 1985 Value Start Value Start Value
Year Sold – 1985 End – Start Year
Comparison of HRS/AHEAD Regressions With AHS Regressions
134
Refereed Papers
Rodda and Patrabansh
when compared with the appreciation rates of the near-elderly and the middle-aged homeowners
than when compared with the appreciation rates of the general population of homeowners.
11
Given that house value appreciation can have both location and time variations, some differences
in the magnitude of AHS and HRS coefficients can be expected because HRS has a different and
AHEAD = Assets and Health Dynamics Among the Oldest Old Study. CAGR = compound annual growth rate. HRS = Health
and Retirement Study. TICS = Telephone Interview for Cognitive Status.
* Indicates signicance at the 5-percent level.
** Indicates signicance at the 1-percent level. The observation level of the sample in this report is a primary home. Some
households have as many as three primary homes in the survey period from 1992 through 2002. Only single-family, nonfarm,
nonmobile, and noncondominium owned primary homes are considered in the sample, which is also conned to homes
with nonimputed and nonmissing house values reported by respondents in both start and end years. CAGR, the dependent
variable, is a measure similar to annualized difference in natural logs of end and start values. Results are weighted to make
inference on the U.S. population of the same age, gender, and race/ethnicity prole as that of the HRS/AHEAD sample using
household weights provided by HRS. All dollar amounts are adjusted to 2002 using the nonseasonally adjusted Consumer
Price Index excluding shelter.
Source: 1992 to 2002 HRS/AHEAD
Respondent older than
74 years in start year
– 0.0121 0.0028** – 0.0097 0.0029** – 0.0098 0.0029**
Interval between end
and start years
– 0.0027 0.0004** – 0.0027 0.0004**
Suburban location of
home
– 0.0060 0.0023** – 0.0060 0.0023**
Rural location of home – 0.0062 0.0024** – 0.0063 0.0024**
Liquid assets indicator 0.0082 0.0038* 0.0082 0.0038*
TICS score less than 5 – 0.0060 0.0078
TICS score missing 0.0039 0.0047
Mexican-Hispanic
respondent
– 0.0117 0.0060 – 0.0117 0.0060
Other Hispanic
respondent
– 0.0057 0.0069 – 0.0056 0.0069
Constant 0.0218 0.0010** 0.0607 0.0159** 0.0603 0.0156**
Fixed effects None Division x
end year
Division x
end year
Other covariates None Yes Yes
N 7,903 7,903 7,903
R
2
0.00 0.08 0.08
Exhibit 4
Covariate
(11) (14) (16)
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Regressions of Compound Annual Growth Rates of House Values
11
In addition, the American Housing Survey (AHS) maximum observation period for a home is 16 years, between 1985
and 2001, while the Health and Retirement Study (HRS) maximum observation period is 10 years, between 1992 and
2002. A large number of homes in the HRS are observed for shorter periods than 10 years and owners may not adjust their
perceptions of house values in shorter time horizons (HRS) as much as in longer time horizons (AHS).
135
Cityscape
Homeowner Age and House Price Appreciation
shorter timeframe, and its geographic breakdown is not as detailed. The inability of the census
division indicator variables to pick up location variation as much as the MSA indicator variables
shows up as a lower R-squared in the HRS models compared with the AHS models.
Instead of the log difference of house values, we prefer to use the CAGR of house values. The log
difference assumes continuous compounding; CAGR, on the other hand, reports annual growth
rate similar to that of interest rates and is suitable for house value growth rates. Despite these
differences, the two measures are more alike than different in terms of computation.
12
In regression
models (5) and (6), we used CAGR of house values as the dependent variable and the similarity
of annualized log difference and CAGR becomes evident by comparing the coefficients of the age
indicator variable in regression models (4) and (6). They are essentially the same.
Regression model (6) serves as the HRS foundation model for further specification testing. In exhibit 4,
we estimate the same regression without the interactions between the end years and census division
indicator variables in regression model (11). The two coefficients are very close but the R-squared
of regression model (11) is practically 0. The interaction indicator variables do not influence the
age effect but explain some location variations to make the regression a better fit. As more covari-
ates were added to the model, the coefficient of age decreased very slightly in magnitude but its
significance remained strong. The Telephone Interview for Cognitive Status (TICS)
13
score did not
have a significant coefficient and the coefficient on age barely changed from -0.0097 to -0.0098.
The 1.0-percentage-point decrease in annual house value appreciation rate for elderly homeowners
is a lower bound of the estimate. The upper bound is a 1.2-percentage-point decrease in CAGR, as
shown in regression model (11).
Other significant covariates in the regressions included the interval between the start year and the
end year, the suburban and rural location of a home compared with an urban location, the pres-
ence of liquid assets, the TICS score, and the homeowner’s being Mexican Hispanic. The longer
the interval between the start year and the end year, the lower the annual house appreciation is.
Suburban and rural homes have lower annual appreciation than do urban homes, presumably
because urban land values are appreciating more rapidly. Homes owned by individuals possessing
liquid assets have higher annual appreciation. The significance of the Mexican-Hispanic indicator
variable is consistent across models while no other demographic variable is significant. Homes with
Mexican-Hispanic homeowners apparently have lower annual appreciation. This control variable
may be picking up a neighborhood effect that the census division variables are not able to pick up.
A graphical representation of how lower appreciation rates affect the values of homes over 20 years
is shown in exhibit 5. The homeowner types include the following:
12
If a house value is $100,000 for 1992 and $250,000 for 2002, in real terms, the compound annual growth rate is 9.6
percent, while the annualized log difference is 9.2 percent, and the log difference is 91.6 percent. Compare those rates to a
crude percentage change of 150 percent and a percentage change per year of 11. Each approach measures the same change
from $100,000 to $250,000, but the 9.6 percent for the compound annual growth rate emphasizes the effect of simple
annual interest compounding over time.
13
The TICS score is a standard multidimensional measure of cognitive function, including mental acuity and memory. The
values range from 1 to 10 with higher scores meaning better cognitive functioning.
136
Refereed Papers
Rodda and Patrabansh
• An average-aged U.S. homeowner who has house price appreciation that matches the Office of
Federal Housing Enterprise Oversight [OFHEO] House Price Index).
14
• A middle-aged homeowner who first measures the house price at age 50.
• A middle-aged-turned-elderly homeowner who begins measuring the house price at age 65 and
continues tracking house price beyond age 75.
• An elderly homeowner who first measures the house price at age 75.
Exhibit 5 displays a simplified representation with smooth appreciation every year, but this
representation is able to convert the appreciation rate differences into constant dollar amounts.
We start off all four types of households in the first year with homes worth $100,000. The home
owned by an elderly homeowner grows in appreciation at the rate of 1.1 percent every year, while
the home owned by a middle-aged homeowner grows in appreciation at the rate of 2.08 percent
per year. The home owned by a middle-aged-turned-elderly homeowner grows in appreciation at
the rate of 2.08 percent per year until the homeowner reaches age 75; after that point, the home
grows in appreciation at the rate of 1.1 percent per year. The average U.S. home, however, grows
in appreciation at the rate of 2.45 percent per year. The annual house appreciation growth rate of
2.08 percent for the middle-aged group is lower than the annual house appreciation growth rate of
2.45 percent for the average U.S. homeowner.
CAGR = compound annual growth rate. OFHEO = Ofce of Federal Housing Enterprise Oversight.
Exhibit 5
Simulated Appreciation of House Values for Different Age Cohorts
14
Office of Federal Housing Enterprise Oversight (OFHEO) appreciation rates assume no home improvements were made
between the repeat sales. If the home improvement projects were known and controlled for, the index for a constant-quality
home would be somewhat lower.
137
Cityscape
Homeowner Age and House Price Appreciation
At the end of the 20th year, the home owned by an elderly homeowner is worth the least. This
simulation assumes no differences in quality between the homes of the young and old. Typically,
the newer homes of younger households were built more recently and include larger rooms with
more amenities, but in this simulation we exclude those differences. In terms of house value, a home
owned by a middle-aged-turned-elderly homeowner is worth less than that of a home owned by a
middle-aged homeowner. A home owned by an average U.S. homeowner does the best. The differ-
ence in real dollar terms between a home owned by an elderly homeowner and a home owned by
a middle-aged homeowner at the end of the 20th year is almost $25,000. The difference between a
home owned by an elderly homeowner and a home owned by an average U.S. homeowner is more
than $35,000. Our estimate of 1.0 to 1.2 percentage points lower appreciation per year for homes
owned by elderly homeowners is smaller in magnitude than Davidoff’s (2004) estimate of 3 percent-
age points per year, but the decrease in annual appreciation of even 1 percentage point is not at all
trivial when considering a longer timeframe, as shown by our simulation in exhibit 5.
Census Public Use Microdata Sample Data and Models
Census Public Use Microdata Sample data provide another useful source for investigating the
relationship between homeowner age and house value appreciation. Unlike HRS data, which are
longitudinal samples for elderly and near-elderly households, census data provide cross-sections of
the U.S. population every 10 years. The cross-sectional nature of the census data makes it neces-
sary to summarize house values at a geographical level such as the MSA before matching 2 census
years to calculate house value appreciation. Using the census data enables us to verify the general
validity of our HRS results and to determine if the omission of building age and poor reporting of
tenure in the HRS data biases the HRS estimates of house price appreciation.
The data for household heads are extracted from the 1-percent PUMS samples in 1990 and 2000.
We calculated house price appreciation at the national, census division, and MSA levels using the
PUMS geographical identifiers. While matching MSAs in 1990 and 2000, we excluded all MSAs
that were defined differently in 1990 and 2000 and all nonmetropolitan areas. Exactly 105 MSAs
had common boundaries in 1990 and 2000.
15
From the sample of noncommercial, noncondo-
minium single-family detached houses, we drew two types of comparison samples.
The first sample consists of two types of households: the young cohort (50 to 59 years old in 1990
and 60 to 69 years old in 2000) and the old cohort (65 to 74 years old in 1990 and 75 to 84 years
old in 2000). The second sample contains the households of the same age groups, but both groups
are now restricted to only those household heads who had lived at their current address for at
least 11 years in 1990 or at least 21 years in 2000. By restricting the sample to nonmovers, it is
less likely that the appreciation rate calculated between the 2 census years represents a change in
15
The metropolitan statistical areas (MSAs) excluded because of changes in boundaries were those that were growing
rapidly in either population or size. If the boundary growth was due to “suburbanizing” elderly homeowners and those
homeowners as movers did a better job of maintaining their property than the nonmovers, then the Public Use Microdata
Sample regressions would underestimate the house price appreciation. If the growing cities tended to be places with high
supply elasticity and more stable house prices, however, then exclusion of those cities might overestimate the house price
appreciation. More analysis is needed to weigh the balance of these effects.
138
Refereed Papers
Rodda and Patrabansh
mobility (especially to new construction). Therefore, the second sample is called the sample with
restricted tenure, while the first is called the sample with unrestricted tenure.
House Value and Appreciation by Age Categories
The HRS data—a longitudinal sample—follow the same households over 10 years. The census
data are not longitudinal, but an age cohort can be created by the age of the household head. In
this case, two samples are compared. Young homeowners were 50 to 59 years old in the 1990
sample and would be 60 to 69 years old in 2000. The old homeowners were 65 to 74 years old in
the 1990 sample and would be 75 to 84 years old in the 2000. The PUMS is a 1-percent sample,
so few of the 1990 households would also appear in the 2000 sample. We assume, however, that
the medians from the included sample are a fair representation of the households had they been
included in both samples. The restricted tenure sample further limits the cohorts by requiring the
households to have not moved in the past 10 years in the 1990 sample and not moved for the past
20 years in the 2000 sample. In other words, the restricted tenure sample tracks the stayer cohort
by excluding movers and newly constructed homes. The unrestricted tenure sample follows the
same cohort by age but includes movers and new homes.
Median house values and appreciation rates for the cohorts in both the restricted and unrestricted
tenure samples are shown in exhibit 6.
16
At the national level, the restricted tenure sample of stayers
shows that the old cohort did 0.31 percentage point worse than the younger cohort did. By census
division, the West South Central had the highest relative gain, or a 1.9-percentage-points difference,
for the old cohort. The lower panel gives the results for the unrestricted tenure sample. The growth
rate for the old cohort is the same (0.64 percent) as that for the restricted tenure sample, but the
young cohort experienced almost no growth (0.03 percent). As a result, the relative gain for the old
cohort is 0.62 percentage point.
PUMS Model Results
The tabulations of house value appreciation offer limited controls beyond age and place. A linear
regression model can include other control variables for length of tenure, building age, unit size,
demographics, and household income. The PUMS limitation of independent samples means that
the unit-level values are aggregated to the MSA-by-age cohort level. We studied 105 MSAs and
2 age cohorts taken from the PUMS data, so the sample size is 210. Income has been divided by
$10,000 (to make the regression coefficient larger), and the values are in year 2000 dollars. The
main purpose of the regression is to test the hypothesis that elderly homeowners have a lower
appreciation rate for their houses, particularly when the homeowner is 75 years old or older. A
second purpose is to gauge the degree of bias that might be introduced in the HRS results from
omitting building age or length of tenure. The PUMS data enable us to include controls for build-
ing age and length of tenure at the aggregate level. Therefore, by comparing specifications with and
without those controls, we can see how much the coefficient on homeowner age is influenced by
the omission of those correlated variables.
16
The numerous repetitions in exhibit 6 between the restricted tenure and unrestricted tenure samples are not
typographical errors. They are the consequence of imputing the house values to be the midpoint of reported categories
and adjusting the top codes of $400,000 in 1990 and $500,000 in 2000 by a factor of 1.25. The national estimates are the
weighted medians of individual house values and do not use the weighted medians for the divisions.
139
Cityscape
Homeowner Age and House Price Appreciation
Restricted Tenure
d
National 122,098 102,300 89,100 81,385 112,500 95,000 0.95 0.64 – 0.31
Census Division
New England 6,474 214,500 181,500 4,338 162,500 137,500 – 2.74 – 2.74 0.00
Middle Atlantic 18,859 148,500 125,400 12,446 137,500 112,500 – 0.77 – 1.08 – 0.31
East North Central 23,312 89,100 75,900 15,720 95,000 85,000 0.64 1.14 0.50
West North Central 9,712 82,500 62,700 6,946 85,000 75,000 0.30 1.81 1.51
South Atlantic 19,722 95,700 82,500 13,753 95,000 85,000 – 0.07 0.30 0.37
East South Central 8,294 69,300 62,700 5,889 75,000 75,000 0.79 1.81 1.01
West South Central 13,035 75,900 62,700 8,801 65,000 65,000 – 1.54 0.36 1.90
Mountain 5,306 95,700 89,100 3,675 112,500 112,500 1.63 2.36 0.73
Pacic 14,869 247,500 214,500 9,817 225,000 187,500 – 0.95 – 1.34 – 0.39
Exhibit 6
Geographic Area
1990 2000 CAGR
a
Median House Value
b
Median House Value (A) (B) (B) Minus (A)
Number Young
c
Old Number Young Old Young Old
($) ($) ($) ($) (%) (%) (%)
Median House Values and Appreciation by Census Division: Cohort Selection With and Without Restricted Tenure (1 of 2)
140
Refereed Papers
Rodda and Patrabansh
a
CAGR is compound annual growth rate of median house values between 1990 and 2000 for each geographical entity. (B) minus (A) is the difference in CAGRs of the old and young
cohorts.
b
Median house values were calculated instead of mean house values because house values were topcoded. Median house values are in 2000 dollars.
c
The young cohort consists of homeowners who were 50 to 59 years old in 1990 and 60 to 69 years old in 2000. The old cohort consists of homeowners who were 65 to 74 years old in
1990 and 75 to 84 years old in 2000.
d
The sample with restricted tenure is conned to households that had been living at their current address for 11 years or longer in 1990 and 21 years or longer in 2000. The sample with
unrestricted tenure can have any length of tenure.
Note: Sample sizes are the total for all age groups and are unweighted. The median house values and CAGRs are weighted by the household weight provided by the Integrated Public Use
Microdata Series (IPUMS).
Source: 1990 and 2000 IPUMS
Exhibit 6
Median House Values and Appreciation by Census Division: Cohort Selection With and Without Restricted Tenure (2 of 2)
Unrestricted Tenure
d
National 164,242 112,200 89,100 135,675 112,500 95,000 0.03 0.64 0.62
Census Division
New England 8,159 214,500 181,500 6,446 162,500 137,500 – 2.74 – 2.74 0.00
Middle Atlantic 23,063 181,500 125,400 17,398 137,500 112,500 – 2.74 – 1.08 1.66
East North Central 29,893 89,100 75,900 24,394 112,500 95,000 2.36 2.27 – 0.09
West North Central 13,096 82,500 62,700 11,531 95,000 75,000 1.42 1.81 0.39
South Atlantic 28,488 102,300 89,100 25,547 112,500 95,000 0.95 0.64 – 0.31
East South Central 11,172 75,900 62,700 9,768 85,000 75,000 1.14 1.81 0.67
West South Central 18,342 82,500 62,700 15,672 75,000 65,000 – 0.95 0.36 1.31
Mountain 8,011 102,300 95,700 7,892 137,500 112,500 3.00 1.63 – 1.37
Pacic 20,794 247,500 181,500 17,027 225,000 187,500 – 0.95 0.33 1.27
Geographic Area
1990 2000 CAGR
a
Median House Value
b
Median House Value (A) (B) (B) Minus (A)
Number Young
c
Old Number Young Old Young Old
($) ($) ($) ($) (%) (%) (%)
141
Cityscape
Homeowner Age and House Price Appreciation
The full regression results for the restricted tenure sample are presented in the appendix. The
primary focus is on the coefficient for homeowner age, which is summarized in exhibit 7. For the
stayer sample (that is, the cohort with tenure restriction), the age coefficient for the full model is
-0.032. This coefficient means that the houses of homeowners 75 years old or older who lived
in the same house for at least 10 years had an annual appreciation rate that was 3.2 percentage
points lower than that of houses owned by the younger cohort. Omitting the building age variables
increases the elderly discount to -3.4 percentage points. On the other hand, omitting the length
of tenure but including the building age reduces the discount to -2.7 percentage points. Omitting
both tenure and building age reduces the discount to -2.5 percentage points. These findings suggest
that the HRS results may be biased downward (toward 0) by omitting controls for length of tenure
and building age that would capture the depreciation effect, but the size of the bias is modest. In
fact, if the regression includes a simple specification of homeowner age, number of rooms, and
household income, the estimated discount is -2.9 percentage points.
The same series of regressions were estimated on the cohort sample without tenure restriction (that
is, the movers and stayers) and the results are shown on the right half of exhibit 7. As expected,
the elderly discount is smaller when the sample includes movers and new construction, but the
estimate is about -2.3 percentage points. This estimate is about twice as large as the HRS elderly
discount even though the data come from approximately the same timeframe and age groups. The
most important difference is that HRS is a longitudinal data set while PUMS has two independent
cross-sections. A second, potentially important, difference is that HRS omits controls for length
of tenure and building age. Despite these differences, the PUMS specifications that exclude those
variables have relatively little impact on the elderly discount. The simplest specification of age,
house size, and income produces the very same discount of -2.3 percentage points.
Note: Full results of these regressions appear in the appendix.
Source: 1990 and 2000 Census Public Use Microdata Sample
Tenure and building age – 0.032 – 0.023
Tenure, not building age – 0.034 – 0.025
Not tenure, building age – 0.027 – 0.021
Not tenure, not building age – 0.025 – 0.021
Exhibit 7
Controlling For Stayers Movers and Stayers
Discount to Elderly Homeowners (75 Years Old or Older) in House Price Appreciation
Discussion
The main conclusion from the HRS/AHEAD data is that elderly homeowners report lower house
value appreciation than do middle-aged homeowners, as summarized in exhibit 8. Measured in
constant dollars, houses owned by people 75 years old or older appreciate annually at rates 1.0 to
1.2 percentage points lower than those of houses owned by middle-aged people younger than 75.
The larger discount corresponds to regressions that do not control for memory acuity and thus
the age coefficient captures the combined effect. In comparison, using AHS data, Davidoff (2004)
142
Refereed Papers
Rodda and Patrabansh
estimated a discount for elderly homeowners of -2.3 to -3.6 percentage points. A similar regression
model on aggregated PUMS data produced elderly discounts in the range of -2.1 to -3.4 percentage
points. As we saw in exhibit 5, even the smallest of these annual differences will have cascading
impacts on the elderly, their children, their communities, and the governmental and nongovern-
mental institutions that hold or insure their mortgages.
What accounts for the differences in these parameter estimates? Several differences in the data
could account for the HRS age discount being smaller than the AHS age discount. The HRS data
(including the AHEAD, War Babies, and Children of the Depression Age supplements) represent
an older distribution of homeowners than that of the AHS. Based on the ending year, 25.9 percent
of the HRS data consist of homeowners 75 years old and older; the comparable AHS figure is
10.7 percent. The higher concentration of elderly homeowners improves the precision of the HRS
estimates, but the AHS may provide a better representation of the elderly discount relative to the
overall population of homeowners. As shown in exhibit 5, the average house price appreciation for
the overall population is 2.45 percent compared with 2.08 for middle-aged homeowners. Adding
that difference (0.37 percentage point) to our estimate generates an elderly discount in the range
of 1.4 to 1.6 percentage points relative to all homeowners and narrows the difference between the
HRS and AHS results.
A second important distinction is that the spells measured by the HRS, 1992 to 2002 or less, are
generally shorter than the spells measured by the AHS, 1985 to 2001. Not only is the span of
survey years shorter for HRS, but a substantial portion of households in the HRS sample was first
interviewed after 1992 or left the sample before 2002. The HRS models clearly show a negative
coefficient on length of spell from beginning to end. It is possible that, if the HRS spells had been
as long as the AHS spells on average, the age discount for the HRS would have been just as large as
what Davidoff (2004) found in the AHS or we have estimated from the PUMS.
Assuming the findings of an elderly discount are correct, what could explain this phenomenon?
Six alternatives have been considered in the literature. Two explanations suggest genuine behav-
ioral differences:
1. Relative undermaintenance by elderly homeowners leads to accelerated property depreciation.
2. Movers maximize wealth with home improvements while stayers minimize expenditure.
AHS = American Housing Survey. HRS = Health and Retirement Study. PUMS = Public Use Microdata Sample.
Models on HRS data – 1.0 to – 1.2
Models on AHS data – 2.3 to – 3.6
PUMS models with tenure restriction – 2.5 to – 3.4
PUMS models without tenure restriction – 2.1 to – 2.5
Exhibit 8
Model Range of Discount (%)
Summary Comparison of House Price Appreciation Discounts for Elderly
Homeowners
143
Cityscape
Homeowner Age and House Price Appreciation
Another four explanations tend to attribute the apparent discount to omitted variables and
respondent error:
3. Elderly retirees move to elastic supply markets in the South and nonmetropolitan areas, where
less appreciation occurs.
4. Homeowner age is correlated with length of tenure or building age.
5. Higher variance of house values is associated with older houses.
6. Self-reported house values are biased from homeowners being out of the housing market and
being poorly informed about price trends.
Several plausible stories explain the lower house price appreciation for elderly homeowners. The
explanation featured in Davidoff (2004) is that elderly homeowners undermaintain their property
and thus their houses do not appreciate as quickly as those of the average homeowner. Unfortu-
nately, HRS does not ask about maintenance spending per se, but supporting evidence from home
improvement projects is present. Elderly homeowners are significantly less likely than middle-aged
homeowners to do a home improvement or major addition (19.2 percent compared with 26.9).
The average amount spent on home improvement projects is less for elderly homeowners than
middle-aged homeowners ($2,826 compared with $4,084), but the difference is not statistically
significant.
17
It is difficult to determine whether this difference in home improvement spending
is enough to account for the lower house price appreciation. Nevertheless, lower home improve-
ment spending by elderly homeowners fits the story that elderly homeowners invest less in, if not
undermaintain, their housing relative to younger homeowners.
A significant portion of housing subsidies provided by HOME and Community Development Block
Grant funding is devoted to rehabilitating homes owned by low- and moderate-income elderly
people. Federal spending may in part offset the apparent undermaintenance by this homeowner-
ship group, preserving affordable housing both for elderly homeowners and the next generation.
Another measure of declining interest in housing investment is in the ownership of second homes.
Only 8.8 percent of elderly homeowners have a second home compared with 15.2 of middle-aged
homeowners. No significant difference in the average liquid assets exists between elderly and
middle-aged homeowners. Elderly homeowners do have higher average out-of-pocket medical
expenses than do middle-aged homeowners ($2,837 compared with $2,055), but the difference is
not significant whether the $0 cases are or are not included in the averages. Also, medical expenses
as a share of liquid assets are not higher for the elderly.
18
Thus, considering the available liquid
assets, the difference in health spending on average does not seem to be enough to crowd out
maintenance spending.
17
The difference in nonzero home improvement spending (excluding the zeros from the averages) is also not significant.
Fisher and Williams (2006) note that the incidence of maintenance is lower in the Consumer Expenditure Survey data than
in the American Housing Survey (47 percent compared with 77), but the maintenance spending per year is nearly twice as
large ($1,152 compared with $622).
18
This unexpected result of lower medical expense relative to liquid assets of the elderly homeowners may be due to
the higher rate of missing data for the elderly homeowners (26.3 percent compared with 19.7). It might also result from
medical insurance reducing out-of-pocket expenses for medical care.
144
Refereed Papers
Rodda and Patrabansh
As homeowners age, they are less likely to move and they are less likely to have second homes. The
regressions presented here compare the combination of movers and stayers with the stayers alone.
Movers are motivated to keep their home in a marketable condition, but stayers may be more con-
cerned with minimizing expense and enjoying “familiar surroundings as they have always been.”
If preferences shift away from housing investment, then elderly homeowners may permit their
properties to depreciate as a way to extract housing equity without having to move. The PUMS
results show that stayers (or the cohort with restricted tenure) have the largest elderly discounts in
house value appreciation.
Another explanation is that retirees move to housing markets with elastic supply. To the extent that
the South, West, and nonmetropolitan markets are more elastically supplied, this explanation is
somewhat plausible. All the regressions in this article control for location to one degree or another;
however, the regression-adjusted results are not consistent with this story; our results in exhibit 6
appear to show the homes of the elderly appreciating faster than those of others in the more elastic
South Atlantic, East South Central, and West South Central census divisions.
The correlation of the homeowner’s age with building age and length of tenure is supported in
the PUMS data, but omitting those variables seems to have little effect on the size of the elderly
discount. Basing calculations on those results, the omission of building age and tenure from the
HRS models should not greatly affect the estimate of the elderly discount. Elderly homeowners
are not making the investment to offset depreciation. Harding, Rosenthal, and Sirmans (2007)
control for holding period and building age but not for homeowner age, which we have shown to
be important. Adapting their estimation strategy, a logical extension of our research is to estimate
maintenance spending while controlling for homeowner age and building age (using either AHS or
Consumer Expenditure Survey data) and then estimate a repeat-sales model while controlling for
imputed maintenance. This approach would control for the endogeneity of maintenance spending
and incorporate the effect of the homeowner’s age.
The TICS score did not have a significant coefficient or a significant effect on the age variables. We
assumed that mental acuity and memory as measured by the TICS score would be correlated with
market awareness, but there appears to be no relation. As measured in HRS, cognitive function
problems do not affect a homeowner’s estimate of house value. More experimentation with other
cognitive function and health measures in HRS might identify a better measure for mental aware-
ness related to house valuation.
Memory and mental acuity are important for distinguishing whether low house value appreciation
by elderly homeowners is a real phenomenon or the result of downward biased estimations.
Homeowners who have not purchased a house in more than 20 years may not realize how much
their house has increased in value over that time. The evidence from Kiel and Zabel (1999) on the
AHS data is that seasoned homeowners have relatively unbiased self-appraisals, but those results
may not apply to the very aged. In our view, the “poor memory” explanation of low house value
appreciation remains viable and requires more direct evidence before it can be refuted in favor of
alternative explanations.
In summary, two independent studies analyzing three separate data sets—HRS, AHS, and PUMS—
have shown a negative and significant relationship between homeowner age and house value
145
Cityscape
Homeowner Age and House Price Appreciation
appreciation. The elderly discount from the HRS is about half the discount from the AHS or PUMS,
however, and that difference is important to long-run planners, including elderly homeowners. We
lack a definitive explanation for why elderly homes appreciate at a slower rate; several explanations
warrant further investigation. Undermaintenance is a leading contender based on the reduction
in home improvement spending, but the difference in spending is relatively modest and probably
reported with error. The driving force may not be health spending or utilities crowding out main-
tenance but rather the preference of many elderly homeowners with long tenure to extract equity
from their homes without selling them.
146
Refereed Papers
Rodda and Patrabansh
Exhibit A-1
Covariate
CAGR
(22) (23) (24) (25)
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Regressions of House Price Appreciation on Cohorts of PUMS Data With Tenure Restriction (1 of 2)
Head older than 74 years in 1990 – 0.0324 0.0086** – 0.0335 0.0084** – 0.0267 0.0071** – 0.0253 0.0059**
Tenure at home 21 through 30 years in 1990 – 0.0481 0.0327 – 0.0141 0.0232
Tenure at home more than 30 years in 1990 0.0114 0.0277 0.0264 0.0206
Building age 21 through 30 years in 1990 0.0574 0.0444 0.0429 0.0368
Building age 31 through 40 years in 1990 0.0092 0.0368 0.0220 0.0268
Building age 41 through 50 years in 1990 0.0687 0.0460 0.0000 0.0000
Building age more than 50 years in 1990 0.0430 0.0340 0.0000 0.0000
Fewer than 4 rooms in 1990 – 0.1778 0.1002 – 0.1936 0.0985 – 0.1757 0.1001 – 0.1988 0.0983*
6–8 rooms in 1990 – 0.0237 0.0222 – 0.0316 0.0208 – 0.0247 0.0219 – 0.0309 0.0208
More than 8 rooms in 1990 0.0638 0.0293* 0.0689 0.0281* 0.0669 0.0291* 0.0743 0.0279**
Married in 1990 – 0.0026 0.0977 – 0.0432 0.0923 – 0.0468 0.0956 – 0.0844 0.0889
Separated, divorced, or widowed in 1990 – 0.0142 0.1000 – 0.0524 0.0959 – 0.0538 0.0986 – 0.0920 0.0932
Non-Hispanic Black – 0.0522 0.0241* – 0.0423 0.0229 – 0.0485 0.0242* – 0.0390 0.0227
Non-Hispanic other race 0.0079 0.0243 0.0038 0.0232 0.0112 0.0232 0.0091 0.0229
Mexican Hispanic – 0.0795 0.0192** – 0.0802 0.0191** – 0.0788 0.0193** – 0.0771 0.0189**
Other Hispanic 0.0678 0.0438 0.0767 0.0428 0.0678 0.0436 0.0754 0.0429
Household income in 1990 – 0.0911 0.0184** – 0.0897 0.0162** – 0.0946 0.0168** – 0.0968 0.0156**
Appendix
147
Cityscape
Homeowner Age and House Price Appreciation
CAGR = compound annual growth rate. PUMS = Public Use Microdata Sample.
* Indicates signicance at the 5-percent level.
** Indicates signicance at the 1-percent level. The observation level of the sample in this report is a matched metropolitan statistical area (MSA). Only noncommercial, noncondominium
single-family detached houses are considered in the sample. CAGR, the dependent variable, is a measure similar to annualized difference in natural logs of end and start values. Results are
weighted by the sum of weights at the MSA level. Median house values are adjusted to 2000 using the nonseasonally adjusted Consumer Price Index (CPI) excluding shelter expenses, and
median household income is adjusted to 2000 using the nonseasonally adjusted CPI.
Source: 1990 and 2000 Integrated Public Use Microdata Series
Exhibit A-1
Covariate
CAGR
(22) (23) (24) (25)
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Regressions of House Price Appreciation on Cohorts of PUMS Data With Tenure Restriction (2 of 2)
Middle Atlantic 0.0029 0.0060 0.0023 0.0059 0.0030 0.0060 0.0019 0.0059
East North Central 0.0355 0.0070** 0.0346 0.0068** 0.0347 0.0069** 0.0326 0.0067**
West North Central 0.0282 0.0085** 0.0262 0.0083** 0.0270 0.0084** 0.0239 0.0082**
South Atlantic 0.0225 0.0086** 0.0202 0.0080* 0.0224 0.0086** 0.0172 0.0074*
East South Central 0.0183 0.0099 0.0168 0.0094 0.0179 0.0099 0.0140 0.0091
West South Central 0.0207 0.0091* 0.0191 0.0086* 0.0202 0.0090* 0.0156 0.0079
Mountain 0.0357 0.0098** 0.0323 0.0092** 0.0352 0.0097** 0.0289 0.0084**
Pacic 0.0220 0.0074** 0.0218 0.0071** 0.0213 0.0071** 0.0190 0.0066**
Mixed division – 0.0171 0.0386 – 0.0076 0.0381 – 0.0075 0.0385 – 0.0052 0.0381
Constant 0.0526 0.0964 0.1129 0.0886 0.1031 0.0937 0.1547 0.0852
N 210 210 210 210
R
2
0.59 0.58 0.58 0.57
148
Refereed Papers
Rodda and Patrabansh
Exhibit A-2
Covariate
CAGR
(26) (27) (28) (29)
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Regressions of House Price Appreciation on Cohorts of PUMS Data Without Tenure Restriction (1 of 2)
Head older than 74 years in 1990 – 0.0230 0.0072** – 0.0249 0.0070** – 0.0209 0.0065** – 0.0211 0.0059**
Tenure at home less than 11 years in 1990 – 0.0307 0.0551 – 0.0839 0.0395*
Tenure at home 21 through 30 years in 1990 – 0.0587 0.0469 – 0.0425 0.0407
Tenure at home more than 30 years in 1990 – 0.0085 0.0359 – 0.0293 0.0312
Building age less than 11 years in 1990 – 0.0570 0.0631 – 0.0785 0.0489
Building age 21 through 30 years in 1990 0.0632 0.0499 0.0240 0.0418
Building age 31 through 40 years in 1990 – 0.0269 0.0402 – 0.0408 0.0360
Building age 41 through 50 years in 1990 0.0484 0.0528 0.0255 0.0477
Building age more than 50 years in 1990 0.0284 0.0372 0.0107 0.0332
Fewer than 4 rooms in 1990 – 0.1244 0.1063 – 0.1296 0.1053 – 0.1237 0.1045 – 0.1556 0.1061
6–8 rooms in 1990 – 0.0100 0.0242 – 0.0185 0.0226 – 0.0118 0.0233 – 0.0065 0.0223
More than 8 rooms in 1990 0.0705 0.0293* 0.0822 0.0285** 0.0747 0.0282** 0.1120 0.0269**
Married in 1990 – 0.0036 0.1115 – 0.0566 0.1075 – 0.0467 0.1073 – 0.1238 0.1024
Separated, divorced, or widowed in 1990 0.0157 0.1186 – 0.0355 0.1154 – 0.0248 0.1149 – 0.0954 0.1120
Non-Hispanic Black – 0.0504 0.0260 – 0.0448 0.0249 – 0.0463 0.0252 – 0.0265 0.0244
Non-Hispanic other race 0.0176 0.0268 0.0181 0.0259 0.0192 0.0257 0.0321 0.0257
Mexican Hispanic – 0.0564 0.0211** – 0.0640 0.0208** – 0.0572 0.0209** – 0.0611 0.0209**
Other Hispanic 0.0626 0.0393 0.0641 0.0392 0.0652 0.0387 0.0769 0.0393
Household income in 1990 – 0.0889 0.0180** – 0.0932 0.0163** – 0.0893 0.0165** – 0.0977 0.0159**
149
Cityscape
Homeowner Age and House Price Appreciation
CAGR = compound annual growth rate. PUMS = Public Use Microdata Sample.
* Indicates signicance at the 5-percent level.
** Indicates signicance at the 1-percent level. The observation level of the sample in this report is a matched metropolitan statistical area (MSA). Only noncommercial, noncondominium
single-family detached houses are considered in the sample. CAGR, the dependent variable, is a measure similar to annualized difference in natural logs of end and start values. Results are
weighted by the sum of weights at the MSA level. Median house values are adjusted to 2000 using nonseasonally adjusted Consumer Price Index (CPI) information minus shelter expenses,
and median household income is adjusted to 2000 using nonseasonally adjusted CPI information.
Source: 1990 and 2000 Integrated Public Use Microdata Series
Exhibit A-2
Covariate
CAGR
(22) (23) (24) (25)
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Coefcient
Standard
Error
Regressions of House Price Appreciation on Cohorts of PUMS Data Without Tenure Restriction (2 of 2)
Middle Atlantic 0.0054 0.0063 0.0054 0.0063 0.0056 0.0062 0.0030 0.0063
East North Central 0.0358 0.0073** 0.0357 0.0071** 0.0361 0.0071** 0.0348 0.0071**
West North Central 0.0320 0.0086** 0.0302 0.0084** 0.0319 0.0085** 0.0289 0.0084**
South Atlantic 0.0278 0.0087** 0.0260 0.0082** 0.0282 0.0086** 0.0186 0.0076*
East South Central 0.0230 0.0101* 0.0233 0.0096* 0.0236 0.0099* 0.0178 0.0094
West South Central 0.0236 0.0093* 0.0225 0.0087** 0.0246 0.0092** 0.0196 0.0081*
Mountain 0.0456 0.0097** 0.0431 0.0090** 0.0459 0.0095** 0.0370 0.0084**
Pacic 0.0271 0.0076** 0.0259 0.0075** 0.0278 0.0074** 0.0231 0.0071**
Mixed division – 0.0702 0.0373 – 0.0594 0.0372 – 0.0630 0.0369 – 0.0612 0.0377
Constant 0.0509 0.1106 0.1339 0.1022 0.0858 0.1069 0.1510 0.0997
N 210 210 210 210
R
2
0.60 0.58 0.60 0.56
150
Refereed Papers
Rodda and Patrabansh
Acknowledgments
The authors acknowledge the careful reviews provided by Theresa DiVenti, Harold Bunce, and
Edward Szymanoski at the U.S. Department of Housing and Urban Development, Office of
Policy Development and Research. The authors also appreciate the comments provided by the
anonymous referees.
Authors
David T. Rodda is a principal economist at Freddie Mac, although this research was done while he
was a senior scientist at Abt Associates Inc.
Satyendra Patrabansh is an associate at Abt Associates Inc.
References
Capozza, Dennis R., Ryan D. Israelsen, and Thomas A. Thomson. 2005. “Appraisal, Agency and
Atypicality: Evidence from Manufactured Homes,” Real Estate Economics 33 (3): 509–538.
Davidoff, Thomas. 2004 (February). Maintenance and the Home Equity of the Elderly. Working
paper. Haas School of Business.
DiPasquale, Denise, and C. Tsur Somerville. 1995. “Do House Price Indices Based on Transacting
Units Represent the Entire Stock? Evidence from the American Housing Survey,” Journal of Housing
Economics 4: 195–229.
Fisher, Jonathan D., and Elliot D. Williams. 2006 (December). Home Maintenance and Investment
Decisions. Working paper. Bureau of Labor Statistics.
Goodman, Allen C., and Thomas G. Thibodeau. 1997. “Age-Related Heteroskedasticity in Hedonic
House Price Equations: An Extension,” Journal of Housing Research 8 (2): 299–317.
———. 1995. “Age-Related Heteroskedasticity in Hedonic House Price Equations,” Journal of
Housing Research 6 (1): 25–42.
Goodman, John L., Jr., and John B. Ittner. 1992. “The Accuracy of Home Owners’ Estimates of
House Value,” Journal of Housing Economics 2: 339–357.
Harding, John P., Stuart S. Rosenthal, and C.F. Sirmans. 2007. “Depreciation of Housing Capital,
Maintenance, and House Price Inflation: Estimates from a Repeat-Sales Model,” Journal of Urban
Economics 61 (2): 193–217.
Heiss, Florian, Michael Hurd, and Axel Borsch-Supan. 2003. Healthy, Wealthy, and Knowing
Where to Live: Predicted Trajectories of Health, Wealth and Living Arrangements Among the
Oldest Old. Working paper 9897. National Bureau of Economic Research.
Hurd, Michael D. 1990. “Issues and Results from Research on the Elderly, Economic Status,
Retirement, and Savings,” Journal of Economic Literature 28: 565–637.
151
Cityscape
Homeowner Age and House Price Appreciation
Kiel, Katherine A., and Jeffrey E. Zabel. 1999. “The Accuracy of Owner-Provided House Values:
The 1978–1991 American Housing Survey,” Real Estate Economics 27 (2): 263–298.
Venti, Steven F., and David A. Wise. 2001a. “Aging and Housing Equity.” In Innovations for
Financing Retirement, edited by Zvi Bodie, F. Brett Hammond, and Olivia S. Mitchell. University of
Pennsylvania Press, Pension Research Council.
Venti, Steven F., and David A. Wise. 2001b. Aging and Housing Equity: Another Look. Working
paper W8608. National Bureau of Economic Research.
Additional Reading
Nelson, Peter B. 2004 (December). Nonearnings Income Migration in the United States: Anticipat-
ing the Geographical Impacts of Baby Boom Retirement. Working papers CRR WP 2004–31,
Boston College, Center for Retirement Research.
Rodda, David T., Ken Lam, and Andrew Youn. 2004. “Stochastic Modeling of Federal Housing
Administration Home Equity Conversion Mortgages with Low-Cost Refinancing,” Real Estate
Economics 32 (4): 589–617.
U.S. Department of Housing and Urban Development (HUD). 2003 (May). Refinancing Premium,
National Loan Limit, and Long-Term Care Premium Waiver for FHA’s HECM Program. U.S. Depart-
ment of Housing and Urban Development, Office of Policy Development and Research.
———. 2000 (May). No Place Like Home: A Report to Congress on FHA’s Home Equity Conversion
Mortgage Program. U.S. Department of Housing and Urban Development, Office of Policy Develop-
ment and Research.
———. 1999. Housing Our Elders. U.S. Department of Housing and Urban Development, Office of
Policy Development and Research.
152
Refereed Papers