Centre for Urban Economics and Real Estate Working Paper 2014– 01 Reverse Mortgage Demographics and Collateral Performance Thomas Davidoff* February 25, 2014 * Assistant Professor, Strategy and Business Economics Division Sauder School of Business, University of British Columbia 2053 Main Mall, Vancouver, BC, V6T 1Z2, Canada. Tel: (604) 822-8325, Fax: (604) 822-8477. Email: [email protected] We are grateful for the support of the Real Estate Institute of British Columbia and the Real Estate Foundation of British Columbia through their contributions to The UBC Centre for Urban Economics and Real Estate.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Centre for Urban Economics and Real Estate
Working Paper
2014– 01
Reverse Mortgage Demographics and
Collateral Performance
Thomas Davidoff*
February 25, 2014
* Assistant Professor, Strategy and Business Economics Division
Sauder School of Business, University of British Columbia
2053 Main Mall, Vancouver, BC, V6T 1Z2, Canada.
Tel: (604) 822-8325, Fax: (604) 822-8477.
Email: [email protected]
We are grateful for the support of the Real Estate Institute of British Columbia and
the Real Estate Foundation of British Columbia through their contributions to
The UBC Centre for Urban Economics and Real Estate.
Reverse Mortgage Demographics and CollateralPerformance
Thomas DavidoffSauder School of Business, University of British Columbia
February 25, 2014
Abstract
Home Equity Conversion Mortgage (HECM) data seem to confirm two concernsabout these federally insured loans offered to older US homeowners. First, originationsare rare, consistent with a familiar disinterest in extracting home equity through saleamong older owners, even those with low wealth. Second, moral hazard and adverseselection appear to operate on HECM’s implicit home price insurance. Demograph-ics mitigate both concerns. Consistent with greater demand among those with lowwealth, HECM loans are more common, more responsive to price appreciation, andmore intensively used in neighborhoods where large fractions of homeowners are blackand Hispanic, and where incomes and property values are below metropolitan averages.The correlation between minority share of homeowners and late-2000s home price bustsexplains most observed selection into HECM on price appreciation within metropoli-tan areas. Selection on price movements and demographics explains away roughly halfof poor collateral performance in HECM loans that has been attributed elsewhere tostrategic undermaintenance.
Keywords: Mortgages, Housing Demand, Social Security and Pensions, PortfolioChoice, Insurance. JEL Classification: G21, R21, H55, G11, G22
1 Introduction
Home Equity Conversion Mortgages (HECMs) offer US homeowners over 62 cash or lines
of credit. Borrowers may defer repayment until they move or die, and their liability at
loan termination is the lesser of the mortgaged home’s net resale value or the outstanding
loan balance.1 The Federal Housing Administration (FHA) provides HECM lenders with
1Industry participants express mixed opinions as to whether shortfalls trigger adverse credit score eventswhen the loan terminates while a borrower is alive. There is no adverse credit event if the loan terminateswith the death of the borrower(s).
1
insurance against shortfalls at termination between collateral value and the balance due, in
exchange for fees due at loan origination and interest charges through the life of the loan.
Insurance fee schedules and allowable loan to value ratios do not vary across markets, nor
do they incorporate time series risk factors such as rent to price ratios. The limited liability
feature is thus likely to be overpriced in some markets at some dates and underpriced in
others.2
HECM data seem to confirm economists’ concerns about the potential size of a market
for home equity loans to older Americans and about moral hazard and adverse selection
problems in the design of FHA HECM insurance. The stated intent of FHA intervention
into the reverse mortgage market was to help the large number of retired homeowners with
modest income to remain in their homes during retirement.3 The fact that only roughly 2%
of eligible homeowners participate (Consumer Financial Protection Bureau (2012)) appears
consistent with the familiar disinterest among healthy older homeowners, even those with
low wealth and incomes, in liquidating home equity through sale or forward mortgages.
Writing near the inception of the HECM program, Venti and Wise (1989) and Feinstein and
McFadden (1989) found low rates of mobility among healthy older homeowners generally in
US panel data, and found little or no evidence that low wealth or illiquidity increases the
propensity to liquidate home equity through selling a home.4 Both analyses conclude that
these facts suggest that the reverse mortgage market is likely to remain small.
FHA has lost money guaranteeing HECM loans, in large part because HECMs were
originated disproportionately near the recent price cycle peak in metropolitan areas that
experienced large subsequent price declines.5 Shan (2011) and Haurin et al. (2013) observe
2Fees have risen and insurable loan to value ratio formulas have shrunk in the wake of the housing crisis.There are market-specific caps on loan size that limit loan to value ratios among higher value properties, andloan to value varies with the age of the borrower (the younger spouse if a couple). Davidoff (2012) presentssimulations that show the implicit put option in lines of credit to be underpriced in many states near thehome price cycle peak, under different assumptions about future price and interest rate movements.
3HECM was enabled by the 1987 Housing and Community Development Act. See Mayer and Simons(1994) and Kutty (1998) for early estimates of the potential market size.
4Venti and Wise (1989) and Venti and Wise (2001) do find that conditional on sale, poorer homeownersare relatively likely to downsize.
5See Integrated Financial Engineering, Inc. (2013) for loss estimates and Davidoff and Wetzel (2013) for
2
that this geographic and temporal selection is consistent with Akerlof (1970)-type exploita-
tion of unpriced variation in the value of the limited liability provision. For example, it is
possible that the reason HECM demand exploded in the “Sand States” (Arizona, California,
Florida, and Nevada) relative to elsewhere in the mid-2000s was that homeowners viewed
markets in these states as the likeliest to see sufficient price declines to generate a positive
expected value for HECM’s implicit home price insurance feature. A new result, presented
in Section 2.5 using Zip Code-level price index data from Zillow, is that there was also sig-
nificant adverse selection on ex-post price performance within metropolitan areas during the
home price boom.
Consistent both with adverse selection across neighborhoods within metropolitan areas,
and with moral hazard on maintenance, Capone et al. (2010) report that homes backing
HECM loans have appreciated at a lower rate than metropolitan averages. Undermainte-
nance and procyclical terminations are now embedded in federal actuarial modeling described
in Integrated Financial Engineering, Inc. (2013). Shiller and Weiss (2000) and Miceli and
Sirmans (1994) highlight the incentive that HECM borrowers have to undermaintain their
homes when potential loan balances exceed collateral value.
This paper shows that HECM appears to have relatively strong appeal at the lower end
of the wealth distribution. In particular, the ratio of HECM loan originations divided by
estimated eligible homeowners, the sensitivity of originations to price increases, and use of
lines of credit conditional on taking on HECM, are significantly larger in neighborhoods with
high minority shares, low property values, and high poverty rates relative to metropolitan
area means. As suggested in Mian and Sufi (2009), and documented in Section 2, neigh-
borhoods with high minority population shares and lower than average property values saw
larger than average price declines during the recent home price bust. This suggests that
some of the poor collateral performance of HECM loans relative to metropolitan area means
may be driven not by older homeowners consciously exploiting default options embedded in
a discussion of the role of geographic and time series variation in originations in generating insurance losses.
3
HECM, but instead by a more innocuous incidental correlation through demographics.
A glance at HECM loan data suggests a strong correlation among race, property val-
ues, and HECM originations. Figure 1 plots the fraction of homeowners who are black or
Hispanic as of the 2000 US Census and the ratio of HECM loan originations between pro-
gram inception in 1989 through 2011 to the number of homeowners over 62 as of 2010 by
Zip Code for three metropolitan areas: Chicago; Los Angeles; and Washington, DC. Racial
location patterns in all three markets are familiar: blacks are predominant in the South and
East of Chicago; in the South and generally away from the Pacific Coast in metropolitan
Los Angeles; and in the North and East of metropolitan Washington, DC. Property values
are lower in these neighborhoods than elsewhere in these generally expensive metropolitan
areas. The spatial pattern of HECM originations looks remarkably similar to the distribu-
tion of minority homeowners. Section 2.2 shows that the graphical correlation in Figure 1
extends nationally. Within metropolitan areas, HECM originations are more common, and
more sensitive to price appreciation, in neighborhoods where property values are lower than
average, and where a larger than average share of homeowners are black or Hispanic.6
Black and Hispanic homeowners are less wealthy than others, so the relative prevalence of
HECM loans in minority neighborhoods suggests that demand for home equity extraction is
negatively correlated with wealth and income. For example, I find in the 2002 of the Health
and Retirement/AHEAD study that the median ratio of non-housing wealth to home equity
among homeowners identifying as either black or Hispanic and aged 70-75 is .13, median
home value is $75,000, and the median ratio of conventional mortgage debt to home value
is .16. For 70-75 year old homeowners who are not black or Hispanic, the medians are 1.05,
125,000 and .08.
A correlation between low wealth and HECM originations would be consistent both with
6One channel through which minority population shares might affect lending warrants remark: it has beenargued that legislatively imposed affordable housing goals may have fed problematic lending practices in theforward mortgage market. Ghent et al. (2013) provide a critical review and empirical contribution. Suchincentives should not have affected HECM loans directly, which are not reported under the Home MortgageDisclosure Act. There could have been indirect effects through capital gains pressure on originations orexpansion of originator presence.
4
program goals and with theoretical life cycle considerations laid out by Artle and Varaiya
(1978). Housing demand is generally deemed increasing in wealth, but with an elasticity less
than one,7 so low non-housing wealth is correlated empirically with both low housing wealth
and a high ratio of home equity to other wealth. For example, I find in the 2002 wave of
HRS/AHEAD, among home owning households age 70 to 75, that the correlation between
non-housing wealth and the ratio of non-housing wealth to primary residence equity is .61.
Intuition and calibrations (De Nardi et al. (2010), Lockwood (2011)) suggest that bequests
are superior goods. For poorer households, then, proceeds from any sale while alive thus
are typically a larger fraction of resources, and wealth taken from the estate presumably
relatively less valued relative to consumption while alive. Thus the ability to to borrow
against proceeds from a sale while alive or from the value of housing left to heirs should
benefit low wealth households more than higher wealth households. The same could be said
about the benefit of selling a home earlier in retirement, though, and there is little evidence
that selling homes has greater appeal at the low end of the income and wealth distribution.
Medicaid rules provide reason to expect that HECM demand would be more skewed
towards the bottom end of the income and wealth distribution than the sale of primary
residences. Medicaid covers health expenditures for poor retirees, and its coverage of the
significant risk of long-term care looms large in the economics of retirement, and home equity
extraction in particular (see Davidoff (2009), Greenhalgh-Stanley (2011) and Nakajima and
Telyukova (2013)). Medicaid places a high tax on non-housing income and assets, but allows
retirees and sometimes their heirs to live in their homes and even retain home equity untaxed
after funding a nursing home stay. Some states impose liens on the homes of Medicaid
recipients, but others do not. Thus converting home equity to cash through sale when a
Medicaid stay is likely or has occurred is costly. By contrast, whether HECM borrowing
is implicitly taxed or subsidized by Medicaid is ambiguous, as outlined in Davidoff (2012).
The prospect or presence of a lien may make a HECM line of credit more attractive, since if
7Davis and Ortalo-Magne (2011) provide a review of the literature and a frequently cited counter-argument.
5
not spent before a move out of the home or death, home equity may be taken by Medicaid.
HECM proceeds cannot be saved, but unused credit is not subject to Medicaid recapture
and credit may be used over time to stay under spending caps imposed by Medicaid.
In Section 2 I make some effort to disentangle race from wealth and income as sources of
HECM demand. Due to data and identification problems, the effort is necessarily incomplete.
A full picture of the relationship between propensity to consume home equity through HECM
and wealth, income, or liquidity would require detailed data on the wealth and incomes of
HECM users. Unfortunately, given the small fraction of the eligible population that uses
HECM, such relationships cannot be gleaned from surveys such as HRS/AHEAD. HECM
loan data offers detail on mortgaged properties, but nothing other than age and gender
or marital status regarding borrower characteristics.8 Merging HECM data with census
aggregate statistics at the Zip Code level enables descriptions of relationships among HECM
originations per eligible household, ethnic and racial population shares, and income-based
measures of poverty rates among older homeowners measures for the elderly. Data on non-
housing financial assets are not available. Naturally there is considerable heterogeneity
of wealth and income among blacks and Hispanics, and the correlation between measured
ethnicity and resources presumably varies across locations. Even if some of the difference in
takeup rates across neighborhoods with different black and Hispanic shares of homeowners
were explained away by available measures of wealth, wealth is in part a result of demand
for savings, which appears to vary across race and Hispanic status conditional on observable
characteristics.9
8Data on borrower incomes and assets was recorded early in the life of HECM, but since this informationis not part of underwriting criteria (all that is required is that no borrower be under 62, and pre-existingmortgage debt be less than available loan proceeds), Rodda et al. (2000) observed many missing valuesand deemed the data unusable, while noting low use among non-whites. There is no current data on thedistribution of income or non-housing assets among borrowers. Redfoot et al. (2007) describe results froma survey that oversamples borrowers, but notice that the survey responses come disproportionately fromnon-Hispanic whites. Readers of an earlier version suggested that minority shares might be particularlylarge among single women. I did not find a strong effect in a cursory look.
9See, for example, Blau and Graham (1990) and Altonji and Doraszelski (2005). Kain and Quigley (1975);Canner et al. (1991); Gabriel and Painter (2003); and Bayer et al. (2004), among many others discuss theidentification of purely racial and ethnic characteristics versus financial factors in housing demand andmortgage outcomes.
6
Academic researchers do not have access to the resale performance of HECM collateral
that Capone et al. (2010) and Integrated Financial Engineering, Inc. (2012) use to find
that mortgaged homes have seen larger price declines that metropolitan averages. However,
it is possible to observe whether the lender claimed a shortfall between the home’s resale
value and the outstanding loan balance at termination. Consistent with the results of FHA
researchers, a large number of insurance claims arise when adjusting the initial appraised
value by the change in the FHFA repeated sale index for the properties’ metropolitan area be-
tween origination and termination would imply no shortfall. Section 2.6 shows that between
one-third and one-half of these shortfalls disappear when Zip Code and low-value property
indexes from Zillow replace FHFA metropolitan indexes in the calculation of baseline market
appreciation.
Recognizing the geographic distribution of HECM loans within metropolitan areas at-
tenuates the scope for strategic exploitation of default options through undermaintenance to
explain poor collateral performance. However, this geographic distribution could have arisen
through “lemon” selling among older homeowners. Regressions presented in Section 2.5 show
that most of the within-metropolitan adverse selection on price declines can be explained by
the appeal of HECM in neighborhoods where minority homeowners and inexpensive homes
are concentrated. The remainder can be explained by lagged home price appreciation. These
facts do not quite imply that adverse selection on price declines was caused by differences in
demographic propensity to use HECM. However, any conscious exploitation of differences in
the expected value of the limited liability feature of HECM were apparently driven chiefly
by differences in predictable changes to supply and demand factors highly correlated with
demographics (e.g. non-prime loan growth).
7
2 Empirical Analysis of HECM Loans
The discussion in the Introduction motivates three sets of empirical questions. First, we
wish to know if Figure 1 generalizes: are local differences in HECM market penetration and
borrower credit use associated with differences in race and observable measures of poverty?
Second, to what extent can demographics explain the relationship between HECM origina-
tions in the mid-2000s and the magnitude of the local housing bust? Third, to what extent
can the underperformance of HECM collateral relative to metropolitan home price index
performance be explained by within-metropolitan adverse selection?
2.1 Data
Table 1 presents summary statistics at the Zip Code and individual level from: FHA HECM
loan-level data, 2000 and 2010 Census data, and Zillow and FHFA repeated home price
indexes. Summary statistics are presented for variables of primary interest in the cases
where FHFA and Zillow data overlap. This is a more urbanized sample than would be a
sample representative of the US as I consider metropolitan price data, and the Zillow data
appears to be available unequally across states. I measure market penetration as the ratio of
HECM originations to eligible homeowners over some period. Log market penetration might
be easier to fit with characteristics, but I include in the analysis the non-trivial number of
Zip Codes with no HECM originations in particular sub-periods of interest.
FHA provides public access to data on each HECM loan originated through mid-2011.
This data includes the appraised value of the home at origination, the borrower or borrowers’
age and gender or marital status; the date of origination (and termination if any); credit
use by year of the loan’s life; the mortgaged home’s Zip Code (and state, county, and
metropolitan area); and an indicator for whether the loan terminated with a balance in
excess of realized property value net of selling costs. From the credit draws and formulaic
interest charges (most HECM loans accumulate interest monthly at a spread over the one-
8
year treasury or LIBOR yield), I estimate the outstanding balance at termination.10
The Decennial Census provides at the level of “Zip Code Tabulation Area” (an approx-
imation of a Zip Code based on a collection of block groups) counts of homeowners by
age, by race and Hispanic status, and by age and poverty status. Combining the Census
and FHA data, we can observe conditional and unconditional correlations among Zip Code
demographic characteristics and Zip Code HECM take-up rates.
Unfortunately, the Census does not provide direct measures of non-housing asset wealth.
Low Zip Code property values could suggest both low financial wealth and low ratios of
housing to other wealth. Since most micro studies indicate wealth and price elasticities
of housing demand below one, we might reasonably expect that high property values are
more likely to signal a high ratio of housing wealth to other wealth using cross-metropolitan
variation than within-metropolitan variation. Given demand for HECM is concentrated in
low property value census tracts, I focus on Census estimates of the 25th percentile of home
values as of 2000. Similar results apply with median values. 2000 is a natural baseline Census
year, as approximately 95% of HECM loans were originated after 1999 (through mid-2011,
the median origination year was 2007).
The Federal Housing Finance Agency (FHFA) publishes a repeated sale home price index
at the metropolitan level. Zillow publishes an index of representative home prices for all
homes and for homes in the upper and bottom tercile of metropolitan home values both at
the metropolitan and at the Zip Code level. At the HECM loan level, where data is available,
multiplying appraised value at origination by the ratio of the FHFA or Zillow index at the
date of termination to the value at origination yields an estimate of terminal collateral value
(before selling costs) that can be compared to the outstanding loan balance.
2.2 HECM Penetration Rates and Demographics
Tables 2 and 3 present regressions of the form:
10Draw data is annual, but interest accrues monthly. Given that most loans feature very large initialdraws, I assume for calculations that all draws are made in the first month of a loan year.
9
Hzt
N̂zt
=α + βminorityz2000 + γ1povertyz2000 + γ2valuez2000 (1)
+∑m
δzmηm + xztµ+ φ∆pzt + εzt
(2)
In specification (1), Hzt is the number of HECM loans originated in period t in Zip Code z.
I estimate the eligible population N̂zt throughout as the number of homeowners over age 64 as
of the 2010 Census. Recognizing that this approximation in part reflects growing population,
I control for the log growth of older population between the 2000 and 2010 censuses. Not all
household heads over 65 who own a home are HECM-eligible: those with spouses under 62
or with conventional mortgage debt above HECM loan limits are not eligible. In estimating
the numerator H, I exclude HECM loans labeled refinances of existing HECMs and those
flagged as used to purchase a home. The controls x also contain the fraction of homes that
are single family in z because HECM rules limit lending to apartment owners.
The poverty measure is the fraction of homeowners over age 64 that were deemed below
poverty income levels in 1999. “value” is the 25th percentile of owner-assessed value as of
the 2000 census. δzmηm is the product of an indicator and a coefficient associated with Zip
Code z lying within metropolitan area m. “minority” reflects the fraction of homeowners
identifying as black or Hispanic as of the 2000 Census. The relationship between HECM
originations and the black and Hispanic shares are naturally different, but these differences
appear to vary across locations, so I adopt the parsimonious lumping into “minority.” ∆pzt
is Zip Code-specific growth in home prices in z as measured by Zillow over some period; in
one specification, I interact this measure with demographic characteristics.
The estimated regression coefficients reflect partial correlations at the neighborhood level.
These coefficients may be quite different from the causal effects of right hand side variables
on demand for HECM loans for several reasons. Among these reasons are the fact that we do
10
not have controls for non-housing wealth, so the effect of minority population share is likely
overstated. Moreover, loan originators may have allocated marketing resources differentially
to minority and poor neighborhoods. Recognizing that lending policies or personnel may be
correlated within metropolitan areas, I cluster standard errors at that level.11
Table 2 presents regression estimates of the ratio of HECM originations over the full
period 1989-2011 to 2010 eligible population on 2000 Zip Code characteristics. The most
consistent result is that there is a significantly positive relationship between HECM origina-
tions per approximate eligible homeowner and the fraction of homeowners (of all ages) with
a respondent that identifies as black or Hispanic. The estimated effect of moving from zero
minority to 100% minority drops significantly from roughly a 7.5% increase in penetration
to roughly 5% when metropolitan area dummy variables are introduced. Either is a very
large effect, given that the sample mean of the dependent penetration ratio is roughly 1.5%
as indicated in Table 1. The roughly offsetting significant coefficients of estimated 2010 eli-
gible owners (negative) and 2000 eligible owners (positive) suggests positive changes in elder
population do not drive originations, or may simply be a symptom of approximation error
in the denominator of the dependent variable.
We find in specification (1) that conditionaly only on Zip Code population and single
family share, the fraction of owners over 64 as of 2000 who are poor is positively and signif-
icantly associated with HECM originations as a fraction of 2010 estimated eligible owners.
The coefficient on older homeowner poverty switches sign and significance conditional on the
minority share, and then becomes indistinguishable from zero conditional on the log 25th
percentile home value.
The coefficient on log 25th percentile home values flips from a positive and significant sign
without metropolitan area fixed effects to significantly negative conditional on metropolitan
area dummies. This result is consistent with the level of home prices and low wealth both
positively influencing originations: given a price elasticity of demand for owner housing less
11Metropolitan dummies with no clustering may not be sufficient to capture phenomena such as a particularlender’s propensity to market in black neighborhoods, for example.
11
than one, we expect to find higher home values in expensive metropolitan areas holding
income and wealth constant. Within metropolitan areas there are theoretically offsetting
wealth and housing to other wealth ratio effects of home values as described above.
Specifications (5) and (6) of Table 2 repeat specification (4), but confine the sample to
the top 20% of Zip Codes for black and Hispanic population share (5) and bottom 20%
(6). The Zip Codes in specification (6) have less than 4% minority homeowner shares;
those in the top 20% have more than 24%. In both the high and low minority samples,
minority share remains significantly positively associated with originations.12 The rate of
income poverty among owners 65 and above in 2000 is significantly positively associated with
demand in heavily minority neighborhoods, but negatively associated with demand in mostly
non-Hispanic white neighborhoods. By contrast, within metropolitan areas, lower quartile
home prices are significantly negatively associated with HECM originations in largely non-
minority neighborhoods, but not in neighborhoods with larger minority concentrations.
In sum, HECM originations are clearly more common in neighborhoods with large mi-
nority population shares. There is more limited evidence that this relates to income or asset
poverty. Poor whites evidently do not commonly use HECM loans.
2.3 Contemporaneous Appreciation and HECM Originations
Table 3 presents regressions of the relationship between growth in HECM originations and
growth in home prices. Given that HECM demand presumably responds with a lag to home
price growth and given the concentration of growth in prices between 2004 and 2007, I
present regressions of the origination share of eligible population 2004-2007 minus the share
1989-2003 on a measure of the home price boom: the ratio of price in 2006 to 2002. Using
the notation in (1), the dependent variable is thusH2004 through 2007−H1989 through 2003
N̂2010.
Specification (1) of Table 3 yields results very similar to those in specification (4) of
Table 3: the significantly positive relationship between log home price boom and originations
12For the full sample, a linear minority specification has a better fit than log.
12
during 2004-2007 does not affect the positive relationship between origination growth and
minority share or low home values. A log point increase in home prices between 2002 and
2006 relative to the metropolitan mean is associated with an approximately 2% increase in
the growth in originations.
Specification (2) of Table 3 shows that originations are significantly more sensitive to
price growth in minority and inexpensive neighborhoods within metropolitan areas. To the
extent that variation in origination growth may be taken to reflect demand rather than
supply growth, this result illustrates a greater marginal (with respect to price appreciation)
propensity to consume home equity through HECM among minority and low wealth house-
holds. However, there is no significant relationship between income-based poverty rates for
older homeowners and the sensitivity of growth in HECM originations to price growth.
Specifications (3) and (4) of Table 3 show similar results in the heavily minority and
heavily non-minority samples described with respect to Table 2. The interactive effects of
minority share and low home prices with changes in home prices are stronger in highly
minority Zip Codes than in low minority Zip Codes.
Similar results arise to those presented in Table 3 if the ratio of originations to population
2004-2007 is considered in isolation (because originations were generally very low prior to
2004) or if the baseline share subtracted from boom period originations includes both the
pre-boom period 1989-2003 and the period after 2008. This is illustrated in Table 5, which
presents the mean minority share and log median home value by year across HECM Zip
Codes. Let yz be either minority share or log median home value as of 2000, let hzt be the
number of HECMs originated in z in year t, and let Ht be the number of HECMs originated
nationwide in t. Table 5 measures:∑
zyzhzt
Ht. Minority shares peaked and median Zip Code
home value hit a trough around the cycle peak.
13
2.4 Credit Use
Table 4 presents regressions of individual level first-year credit divided by the initial credit
limit for all HECM loans issued prior to 2008. Starting in 2008, a large fraction of loans
required immediate withdrawal of all credit; prior to 2008, the large majority of loans were
lines of credit. The regression controls are: the log appraised value, log credit limit, and
dummies for metropolitan area, (younger) borrower age, and borrower gender or marital
status. Specifications (1) and (2) differ in that the latter excludes loans for which there is
a binding cap, so that the loan to value ratio is constrained by an absolute dollar amount.
Specification (3) limits the sample to loans that can be merged with Zillow Zip Code price
data. In all three cases, we find a significantly positive coeffcients on minority population
share and the fraction of homeowners over 65 with poverty incomes on credit use. High log
25th percentile Zip Code home value and own home values are associated with significantly
less credit use. Log Zillow price relative to January, 2001, is also significantly associated with
more credit use. All effects are highly significant, except for poverty, which has marginal
significance. The results in Table 4 provide a bit more support for a role for low income
per se generating demand for HECM, as opposed race. to an incidental correlation through
race.
2.5 Ex-post Zip Code Price Declines
Davidoff and Wetzel (2013) and Shan (2011) show that HECM originations during the period
of peak home prices in the mid-2000s predict subsequent price crashes across metropolitan
areas. The first specification of Table 6 shows that this is true within metropolitan areas,
too. A one percent difference in originations 2004-2007 divided by 2010 eligible households
in a given Zip Code is associated with a 3% larger price crash as measured by Zillow between
January, 2006 and January, 2011, both relative to metropolitan means. Note a larger ratio
of 2006 to 2011 price implies a larger crash.
Specification (2) of Table 6 adds controls for minority share and log 25th percentile home
14
value. These controls reduce the magnitude of the coefficient of HECM origination share
by roughly half, to 1.5%. Allowing in a crude way for the likelihood that the relationship
between originations and price crash magnitude is likely to be larger where the home price
cycle had greater amplitude, specification (3) shows that when minority share and log 25th
percentile home value are interacted with an indicator for a metropolitan area lying within
the “Sand States” of Arizona, California, Florida, or Nevada, the coefficient on crash falls
again by almost half. Thus the coefficient on HECM market penetration 2004-2007 falls
from approximately 3 to approximately .8 with just two Zip Code controls interacted with
an approximation of crash magnitude. Note that the main effect of being in a Sand State is
captured by metropolitan area dummies in all specifications.
Specifications (4) and (5) of Table 6 show that within the low and high minority share
subsamples, there is no significant relationship of origination share and subsequent price
crash. Specification (6), restoring the full sample, adds to specification (3) a control for
log price appreciation 2002-2006. This control reduces the estimated relationship between
HECM peak period market penetration and ex-post price decline to a value of roughly .2,
statistically indistinguishable from zero.
Evidently, controls for demographics substantially weaken the estimated relationship be-
tween ex-post price crashes and HECM originations. This does not preclude the possibility
that HECM grew in particular Zip Codes because borrowers anticipated limited liability
value there. The fact that controlling for lagged capital gains reduces the coefficient dra-
matically could simply mean that the unpriced information borrowers exploited in choosing
whether or not to use HECM was mostly based on a (correct ex-post) belief that gains dur-
ing the price boom 2002-2006 signalled future price declines. Similarly, the pool of potential
borrowers may have recognized that home values in areas seeing the largest relaxation of
credit constraints were likely to correct dramatically.
15
2.6 Collateral underperformance relative to market
FHA has access to the transaction prices of mortgaged homes under HECM. They (Capone
et al. (2010) and Integrated Financial Engineering, Inc. (2012)) find that relative to FHFA
repeated sales home price indexes, HECM homes appreciate at a lower rate, and more so
when the homes have low appraised value at origination. While there is no public access to
collateral resale values, public FHA data provides a noisy signal of changes in home value.
Public data includes an indicator of whether the loan servicer makes an insurance claim
because the home is worth less than the outstanding balance at termination.13 Credit use
by year of the loan’s life is also public. Given the formulaic interest rate calculation (the
one-year treasury or LIBOR plus a lender’s margin plus FHA interest rate charges for the
mortgage insurance), we can estimate loans’ annual balances.
Table 1 reports summary statistics for individual loan data among HECM loans that have
terminated. Approximately 7.5% of all terminated loans feature a shortfall claim. I exclude
from tabulation loans originated by Financial Freedom, the largest originator in the HECM
data, because Davidoff and Wetzel (2013) reports highly peculiar claim behavior among
these loans. This exclusion does not affect qualitatively the patterns described below.14
Figure 2 plots the fraction of loans that generate a shortfall insurance claim by rounded
estimated ratio of outstanding balance estimated to home value. The solid line presents
the shortfall claim when loan to value ratios are estimated based on appreciation of the
property at the FHFA estimated metropolitan average rate. The dashed line depicts the
rate of shortfall claims by estimated loan to value when home price appreciation is based
on Zillow’s Zip Code level data. The dotted line shows the rate of shortfall claim when
13There are two types of claims. An unusual type, that does not trigger the “shortfall” indicator I consider,occurs when the outstanding loan balance is close to the original appraised value of the home (or maximuminsured value when that is lower due to caps on insured value and loan amounts). This type of claim onlyarises when the loan has been outstanding at a high interest rate long enough, and hence is rare amongHECM loans, of which the median origination year in the data is 2000. I deem a shortfall claim only tooccur when the home is sold for less than the outstanding balance and the insurer requests reimbursement.
14Financial Freedom was a subsidiary of the failed IndyMac Bank, and appears to feature a very largerate of “Type I” errors in which I estimate the loan to have a outstanding balance to value ratio far greaterthan 100%, and yet there is no claim.
16
home price appreciation is based on Zip Code level data for homes in the bottom tercile of
metropolitan home values. There is a substantial gap between the probability of a shortfall
claim when the estimated loan to value ratio is below 100% depending on whether the resale
value at termination is calculated based on the FHFA metropolitan home price index, the
Zillow Zip Code index for all homes, or the Zillow Zip Code index for homes with values in
the bottom tercile of all metropolitan area values.
A shortfall claim when the outstanding balance at is above roughly 90% of true value
may not be surprising, given general undermaintenance among older homeowners (Davidoff
(2004)) and selling costs. Among non-Financial Freedom loans with mark-to-market out-
standing balance to value between 75% and 90% at termination based on FHFA metropolitan-
level price appreciation, I find that 33% feature a shortfall claim. By contrast, only 23% of
such loans feature shortfall claims when home values are marked to market via Zillow Zip
Code level appreciation. When homes are marked to market based on Zillow appreciation
calculated only among bottom tercile homes, the fraction of terminated loans with a shortfall
claim falls to 19.8%.
Given the strong correlation between minority shares and price crashes and the likelihood
that Zip Code level appreciation is measured with error, it is interesting to consider the
shortfall claim rate among homes in Zip Codes with low minority shares. Confining the
sample to terminated non-Financial Freedom loans in neighborhoods with minority shares
below the metropolitan mean and with median home values as of 2000 greater than the
metropolitan mean, the rate of shortfall claims among homes with FHFA marked to market
balance to value ratios between .75 to .9 falls below 19%, from 33% without the sample
limitation. The shortfall claim rate based on balance to Zillow-based value for all Zip Code
homes in the low minority, high value sample is 17.5%. The smaller decline in Zip Code
based claim rates makes sense because that index already incorporates differences in price
declines based on demographics. Interestingly, however, the shortfall rate falls from 19.8%
to 9.6% using the bottom tercile Zip Code price index from Zillow.
17
Summarizing, when market price appreciation is calculated based on FHFA metropolitan
level price growth, we find that almost a third of homes that would have had outstanding
balances between 75% and 90% of home value if they had appreciated at market level ap-
preciation generate shortfall claims. This suggests that owners substantially undermaintain
homes. Between one-third and half of these anomalous shortfall claims disappear when the
benchmark for appreciation is neighborhood level appreciation and when differences in ap-
preciation based on initial home values and minority population shares are recognized. Even
more disappear conditional on low minority shares when the local bottom tercile price index
is the benchmark for market appreciation.
3 Conclusion
HECM loans have been originated more frequently, and with more sensitivity to capital gains,
in neighborhoods with high minority concentrations and property values below metropolitan
averages. Among neighborhoods with high minority concentrations, the rate of poverty
among 65 year olds is also significantly positively associated with high origination rates.
Conditional on taking on a HECM loan, residents of minority, low income, and low
property value neighborhoods, as well as owners of homes below neighborhood averages, use
significantly more credit than others. There is thus reason to believe, contrary to evidence
from the sale of homes, that there is substantially greater absolute and marginal propensity
to consume home equity extraction among lower wealth households than others. Living in
a high property value neighborhood within a metropolitan area is associated with less use
of HECM loans, but living in expensive metropolitan areas and receiving capital gains are
associated with significantly more use. Thus the ratio of home equity to wealth appears to
increase demand for HECM loans. A conjecture is that Medicaid rules, which make owning
a home attractive, but do not necessarily punish use of HECM loans, may explain part of
the difference in appeal at the lower end of the income and wealth distribution between sale
18
and HECM as means of equity extraction.
The same types of neighborhoods in which HECM loans were relatively popular saw
large price declines after 2006. This correlation explains away essentially all of the newly
documented within-metropolitan adverse selection on price, and between one-third and half
of the seeming moral hazard on maintenance suffered by FHA in their HECM insurance
program. Conceivably, this selection into HECM based on minority population share and
relatively low property values could have been driven by a prevailing expectation that the
home price insurance effectively offered by FHA would have most value in these neighbor-
hoods. Life cycle theoretical considerations though, make it highly plausible that causality
ran from liquidity-based demand to an incidental correlation with ex-post price declines. An
interesting possibility mentioned by some industry participants is that errors or fraud in
appraisals might explain some of the residual collateral underperformance.
An ancillary result that warrants further research relates to wealth inequality. Within
metropolitan areas, older homeowners who started the 2000s in neighborhoods with less
housing wealth, more poverty, and presumably less non-housing wealth, saw considerably
larger home price declines after 2007 than other homeowners. Given evidently non-trivial
differences in appetites for spending home equity, and the possibility that changes in the
income distribution will lead to continued underperformance of poorer neighborhoods’ prices
relative to metropolitan area means (pursuant to Gyourko et al. (2004)), the distribution of
resources across Americans of homeowning age may grow more considerably more unequal
through retirement.
References
Akerlof, George A., “The market for” lemons”: Quality uncertainty and the market
mechanism,” The Quarterly Journal of Economics, 1970, 84 (3), 488–500.
Altonji, Joseph G. and Ulrich Doraszelski, “The Role of Permanent Income and De-
19
mographics in Black/White Differences in Wealth,” Journal of Human Resources, 2005,
40 (1), 1–30.
Artle, Roland and Pravin Varaiya, “Life Cycle Consumption and Homeownership,”
Journal of Economic Theory, 1978, 18 (1), 38–58.
Bayer, Patrick, Robert McMillan, and Kim Rueben, “What Drives Racial Segrega-
tion? New Evidence Using Census Microdata,” Journal of Urban Economics, 2004, 56 (3),
514–535.
Blau, Francine D. and John W. Graham, “Black-White Differences in Wealth and
Asset Composition,” The Quarterly Journal of Economics, 1990, 105 (2), 321–339.
Canner, Glenn B, Stuart A Gabriel, and J Michael Woolley, “Race, Default Risk
and Mortgage Lending: A Study of the FHA and Conventional Loan Markets.,” Southern
Economic Journal, 1991, 58 (1).
Capone, Charles A., Karen L. Chang, and Colin A. Cushman, “Identification of
Home Maintenance Risk in Reverse Mortgages: An Empirical Examination of Home Price
Appreciation among HECM Borrowers,” working paper, U.S. Department of Housing and
Urban Development 2010.
Consumer Financial Protection Bureau, “Reverse Mortgages,” Report to Congress, US
Government 2012.
Davidoff, Thomas, “Maintenance and the Home Equity of the Elderly,” Working Paper
03-288, Fisher Center for Real Estate, UC Berkeley 2004.
, “Housing, Health, and Annuities,” Journal of Risk and Insurance, 2009, 76 (1), 31–52.
, “Can ”High Costs” Justify Weak Demand for the Home Equity Conversion Mortgage?,”
Working Paper, Sauder School of Business, UBC 2012.
20
and Jake Wetzel, “Do Reverse Mortgage Borrowers Use Credit Ruthlessly?,” Working
Paper, Sauder School of Business, UBC 2013.
Davis, Morris A. and Francois Ortalo-Magne, “Househould Expenditures, Wages,
Rents,” Review of Economic Dynamics, 2011, 14 (2), 248–261.
Feinstein, Jonathan and Daniel McFadden, “The Dynamics of Housing Demand by
the Elderly: Wealth, Cash Flow and Demographic Effects,” in David Wise, ed., The
Economics of Aging, Chicago: University of Chicago Press, 1989.
Gabriel, Stuart and Gary Painter, “Paths to Homeownership: An Analysis of the Resi-
dential Location and Homeownership Choices of Black Households in Los Angeles,” Jour-
nal of Real Estate Finance and Economics, 2003, 27 (1), 87–106.
Ghent, Andra C., Ruben Hernanandez-Murillo, and Michael T. Owyang, “Did
Affordable Housing Legislation Contribute to the Subprime Securities Boom?,” Working
Paper, Arizona State University 2013.
Greenhalgh-Stanley, Nadia, “Medicaid and the Housing and Asset Decisions of the El-
derly: Evidence from Estate Recovery Programs,” Working Paper, Syracuse University
2011.
Gyourko, Joseph, Christopher J. Mayer, and Todd Sinai, “Superstar Cities,” 2004.
Working Paper, Wharton.
Haurin, Donald, Chao Ma, Stephanie Moulton, and Jason Seligman, “Spatial
Variation in Reverse Mortgages Usage: House Price Dynamics and Consumer Selection,”
Working Paper, Ohio State University 2013.
Integrated Financial Engineering, Inc., “Actuarial Review of the Federal Housing Ad-
ministration Mutual Mortgage Insurance Fund HECM Loans For Fiscal Year 2012,” Tech-
nical Report, For US Department of Housing and Urban Development 2012.
21
, “Actuarial Review of the Federal Housing Administration Mutual Mortgage Insurance
Fund HECM Loans For Fiscal Year 2013,” Technical Report, For US Department of
Housing and Urban Development 2013.
Kain, John F. and John M. Quigley, Housing Markets and Racial Discrimination 1975.
Kutty, Nadinee, “The scope for poverty alleviation among elderly homeowners in the US
through reverse mortgages,” Urban Studies, 1998, 35 (1), 113–130.
Lockwood, Lee M., “Incidental Bequests: Bequest Motives and the Choice to Self-Insure
Late-Life Risks,” Working Paper, NBER 2011.
Mayer, Christopher and Katerina Simons, “Reverse Mortgages and the Liquidity of
Housing Wealth,” Journal Of The American Real Estate And Urban Economics Associa-
tion, 1994, 22 (2), 235–255.
Mian, Atif and Amir Sufi, “The Consequences of Mortgage Credit Expansion: Evidence
from the U.S. Mortgage Default Crisis,” Quarterly Journal of Economics, 2009, 124 (4).
Miceli, Thomas and C.F. Sirmans, “Reverse Mortgage and Borrower Maintenance Risk,”
Journal of the American Real Estate and Urban Economics Association, 1994, 22 (2), 257–
299.
Nakajima, Makota and Irina A. Telyukova, “Reverse Mortgage Loans: a Quantitative
Analysis,” Working Paper, UCSD 2013.
Nardi, Mariacristina De, Eric French, and John Bailey Jones, “Why do the Elderly
Save? The Role of Medical Expenses,” Journal of Political Economy, 2010, 118 (1), 39–75.
Redfoot, Donald L., Ken Scholen, and S. Kathi Brown, “Reverse Mortgages: Niche
Product or Mainstream Solution? Report on the 2006 AARP National Survey of Reverse
Mortgage Shoppers,” Report 2007-22, AARP 2007.
22
Rodda, David T., Christopher Herbert, and Hin-Kin Lam, “Evaluation Report of
FHA’s Home Equity Conversion Mortgage Insurance Demonstration,” Prepared for US
Department of Housing and Urban Development, Abt Associates 2000.
Shan, Hui, “Reversing the Trend: The Recent Expansion of the Reverse Mortgage Market,”
Real Estate Economics, 2011, forthcoming.
Shiller, Robert and Allan Weiss, “Moral Hazard in Home Equity Conversion,” Real
Estate Economics, 2000, 28 (1), 1–31.
Venti, Steven F. and David A. Wise, “Aging, Moving, and Housing Wealth,” in David A.
Wise, ed., The Economics of Aging, Chicago: University of Chicago Press, 1989, pp. 9–48.
and , “Aging and Housing Equity: Another Look,” 2001. NBER Working Paper 8608.
23
Figure 1: Latitude and longitude of Zip Codes in Chicago; Washington, DC; and Los Angeles.Data point radii are proportional to HECM originations divided by homeowners over 64reported in the 2010 US Census (left column), and to black and Hispanic homeowners dividedby all homeowners in the 2010 Census (right column)
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●● ●
●
●
●●
●●
●
●
●
●
●
●●
●
●
●
●
●●● ●
●●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
● ●●
● ●
●
●●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
● ●●
●
●
●
●
●
●
−88.6 −88.4 −88.2 −88.0 −87.8 −87.6
41.2
41.4
41.6
41.8
42.0
42.2
42.4
Chicago HECM penetration
longitude
latit
ude
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●● ●
●
●
●
●
●●●
●
●
●
●
●
●
●
●●
●●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●● ●●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
● ●
●
●●
●
●
●
●
● ●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●●
●●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
● ●●
●
●
●
●
−88.6 −88.4 −88.2 −88.0 −87.8 −87.6
41.2
41.4
41.6
41.8
42.0
42.2
42.4
Chicago minority concentration
longitude
latit
ude
●●●●
●
●●●
● ●●●
●●
●
●
●
●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●●
●
●●
●●
●●●
●●●●
●●
●
●●●
●●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●●
●●●
●●
●●
●●
●
●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
−78.0 −77.5 −77.0 −76.5
38.2
38.4
38.6
38.8
39.0
39.2
39.4
Washington HECM penetration
longitude
latit
ude
●●●●●
●
●
●●●●
●
●●
●
●
●
●●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●●
●
●
●
●●●
●
●
●●
●
●●●●
●
●
●
●●
●
●
●
●●
●
●●
●
●
●●
●
●●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●●●
●●
●●
●
●●
●●
●
●● ●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
−78.0 −77.5 −77.0 −76.5
38.2
38.4
38.6
38.8
39.0
39.2
39.4
Washington minority concentration
longitude
latit
ude
●
●
●
●●
● ●●
●●●
●●●
● ●●
●
●
●●●●
● ●●
●●
●●
●
●
●●
●●
●●
●
●
●●●
●
●
●
●●
●
●
●●
●●
●●
●●
●●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
● ●●●●●●● ●
●
● ●●
●
●
●●
●●●●
●
●
●
●●
●●●
●
●●●●
●
●
●●
●
●●
● ●●
●
●● ●●●●
●
●●● ●
●
●
●
●
●
● ●
●●●● ● ●
●
● ●●
●●●
●
●
●
● ●
●●
●●
●
●
●
● ●
●
●
●
●●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●●●
●●
●●●
●●●●●
●●
●
●●
●●●●●●●
●●
●●
●
●
●
● ●
●
●●●●
●●
● ●
●
● ●
●
●●
●
●● ●●
●●
●●
●
●
●
−118.8 −118.6 −118.4 −118.2 −118.0 −117.8
33.8
34.0
34.2
34.4
34.6
Los Angeles HECM penetration
longitude
latit
ude
●●● ●
●●●
●
●
●
●
●●● ● ●
●
●
●
●●
●●
●
●
●
● ●●
●
●
●
●
●
●
●●
●●
●●●●
●●●●●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●●
●●●
●
●
●●●●●
●
●
●●●
●
●
● ● ●●●
●●
●● ●
●
●●
● ●●●●
●
●●
● ●
●
●
●
●
●
● ●
●●●● ● ●
●
●●
●
●●●
●
●
●
●●
●●
●●
●
●
●
● ●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●●●
●●
●
●●
● ●●●
●
●●●
●●
●●●●●●●●
●
●●
●
● ●
●●●●●
●●
●●
●
●●●
●●
●
●
● ●●
●●
●●
●●
●
−118.8 −118.6 −118.4 −118.2 −118.0 −117.8
33.8
34.0
34.2
34.4
34.6
Los Angeles minority concentration
longitude
latit
ude
24
Figure 2: Default rates by outstanding balance as a percentage of estimated property value,with property estimated from different data sources. Solid line: estimated value at termina-tion based on zip code appreciation from Zillow zip code median sale price index. Dashedline: estimated value at termination based on FHFA metropolitan-level repeated sale index.Dotted line: resale based on Zillow Zip Code index for properties in the bottom tercile ofmetropolitan value.
60 80 100 120
0.0
0.2
0.4
0.6
0.8
1.0
Outstanding balance to mark−to−market value
Fra
ctio
n of
loan
s w
ith a
sho
rtfa
ll cl
aim
FHFAZillow ZipZillow Zip Bottom Tier
25
Table 1: Zip Code Level Summary StatisticsVariable Obs Mean Std. Dev Min Max
Zip Code LevelOriginations 2004-2007Owners 65+ in 2010
6,832 0.016 0.016 0 0.168Originations 1989-2011Owners 65+ in 2010
6,832 0.04 0.03 0.002 0.3Originations 2004-2007 - Originations 1989-2003
Owners 65+ in 20106,832 0.01 0.013 -0.036 0.137
Poverty Rate among owners 65+ in 2000 6,832 0.121 0.083 0 0.657Black+Hispanic share of all owners 2000 6,832 0.156 0.199 0.002 0.986Median Home Value 6,832 150621 96,471 22,000 995,20025th%ile Home Value 6,832 120930 76,138 11,200 795,200Single Family % 6,832 0.872 0.129 0.012 1Homeowners 65+, 2010 6,832 1,428 1,107 31 12,564Homeowners 65+, 2000 6,832 1,262 1,104 14 13,834Log Price 2006/2002 6,832 0.433 0.255 -0.205 1.274Log Price 2006/2011 6,832 0.246 0.263 -0.404 1.206Sand State (CA, AZ, FL, NV) 6,832 0.252 0.434 0 1
First year draw dataappraisal 317,258 238,328 156,463 17,500 999,999credit limit 317258 139,098 68,730 8399 485,957First yr. credit/appraisal 317,258 0.662 0.277 0 1
Terminated Loans DataOutstanding BalanceZip indexed value
91,152 .497 .307 5.74e-07 3.3357Outstanding Balance
FHFA metro indexed value116,111 .495 .271 -2.146 2.537
Outstanding BalanceFHFA metro indexed value
91,654 .501 6.389 -1,931 4.453Shortfall claim 116,111 .075 .263 0 1
26
Table 2: Regressions of Ratio of HECM originations 1989-2010 to homeowners over 65 in2010 on Fraction of homeowners over 65 with below-poverty incomes as of 2000, fraction ofhomeowners who are black or Hispanic, and log median value among owner occupied homes.Specifications (4)-(6) includes metropolitan area dummy variables
( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 )
constant -0.0285* 0.0178** -0.1441** 0.0919** 1e-04 0.0862**( 0.0134 ) ( 0.0074 ) ( 0.0241 ) ( 0.0304 ) ( 0.1288 ) ( 0.0358 )
poverty 0.0585* -0.0553** 0.0068 0.01 0.0558 -0.0284**( 0.031 ) ( 0.0168 ) ( 0.0167 ) ( 0.0163 ) ( 0.0344 ) ( 0.0121 )
single family % 0.0322** 0.0206** 0.0242** 0.0389** 0.0714** 0.0161**( 0.0078 ) ( 0.006 ) ( 0.0059 ) ( 0.0054 ) ( 0.0164 ) ( 0.0051 )
Log 65+, 2000 0.0179** 0.0131** 0.0173** 0.0139** 0.027** 0.004**( 0.0028 ) ( 0.0028 ) ( 0.0025 ) ( 0.0024 ) ( 0.009 ) ( 0.0017 )
Log 65+, 2010 -0.0219** -0.0109** -0.0142** -0.014** -0.0129** -0.0116**( 0.0034 ) ( 0.0039 ) ( 0.0037 ) ( 0.0024 ) ( 0.0053 ) ( 0.0031 )
Log owners, 2000 -0.0143** -0.0099** -0.0104** -0.004 -0.0205* 0.0068*( 0.003 ) ( 0.0031 ) ( 0.0034 ) ( 0.0028 ) ( 0.0112 ) ( 0.003 )
Log owners, 2010 0.0212** 0.0075** 0.0061** 0.0035 0.0099 0( 0.0024 ) ( 0.0025 ) ( 0.0025 ) ( 0.0022 ) ( 0.0069 ) ( 0.003 )
minority 0.072** 0.0751** 0.0463** 0.0544** 0.1695**( 0.0101 ) ( 0.009 ) ( 0.0077 ) ( 0.0086 ) ( 0.061 )
Log 25th %ile value 0.0142** -0.0095** -0.0098 -0.0075**( 0.002 ) ( 0.0027 ) ( 0.0129 ) ( 0.003 )
Adj. R-sq. 0.1 0.26 0.3 0.64 0.58 0.57degrees of freedom 6825 6824 6823 6535 1189 1188Metro Dummies N N N Y Y YMinority Subset NA NA NA NA High Low
Notes: In all regression tables, standard errors clustered at the metropolitan level and *denotes significant at 5%, ** at 1%.
27
Table 3: Regressions of Zip Code HECM originations 2004-2007 minus 1989-2003 dividedby 2010 estimated eligible households on characteristics and Zillow Zip Code price growthestimate January 2002 to January 2006
( 1 ) ( 2 ) ( 3 ) ( 4 )
constant 0.0769** -0.0166 -0.1036* 0.0128( 0.0125 ) ( 0.0171 ) ( 0.0482 ) ( 0.0169 )
minority 0.0175** 0.0139** 0.0093* 0.0101( 0.0039 ) ( 0.0036 ) ( 0.0048 ) ( 0.0528 )
Log Price 2006/2002 0.0217** 0.2518** 0.449** 0.1726**( 0.0036 ) ( 0.0456 ) ( 0.0876 ) ( 0.0622 )
Log 25th %ile value -0.0075** 0.0012 0.0063 -5e-04( 0.0011 ) ( 0.0014 ) ( 0.0039 ) ( 0.0015 )
poverty -0.002 -0.0082 0.0167 -0.0111( 0.0073 ) ( 0.0152 ) ( 0.0322 ) ( 0.0123 )
single family % 0.0086** 0.0079** 0.0168** 0.004*( 0.0023 ) ( 0.0022 ) ( 0.0057 ) ( 0.002 )
Log 65+, 2000 -0.0016* -9e-04 0.0023 -0.0018**( 9e-04 ) ( 8e-04 ) ( 0.0027 ) ( 7e-04 )
Log 65+, 2010 6e-04 5e-04 0 -2e-04( 0.0012 ) ( 0.0011 ) ( 0.0025 ) ( 0.0015 )
Log owners, 2000 0.0022* 0.0021* 0.0011 0.002*( 0.001 ) ( 9e-04 ) ( 0.0038 ) ( 0.0012 )
Log owners, 2010 -0.001 -0.002* -0.0013 -0.0011( 0.001 ) ( 9e-04 ) ( 0.0025 ) ( 0.0017 )
minority × price growth 0.0026** 0.0129** 0.0014( 9e-04 ) ( 0.0039 ) ( 0.0028 )
Log 25th %ile value × price growth -0.0201** -0.0359** -0.0146**( 0.0039 ) ( 0.0079 ) ( 0.0051 )
poverty × price growth 0.0181 -0.0062 0.012( 0.0401 ) ( 0.0779 ) ( 0.0397 )
Adj. R-sq. 0.56 0.59 0.58 0.28degrees of freedom 6534 6531 1185 1184Metro Dummies Y Y Y YMinority Subset NA NA High Low
28
Table 4: Regression of individual loan log ratio of credit used in first year of loan’s lifeto initial principal limit on borrower, loan, and Zip Code characteristics. Specification(1) includes and (2) excludes loans where loan to value ratios are reduced by time- andmetropolitan-varying caps on insured value. Specification (3) further excludes loans withmissing Zillow price data. All three specifications control for dummy variables for age,gender, metropolitan area, and origination year are included. All HECM loans prior to 2008(when fixed rate lump sum loans were initiated) merged with Zip Code data tabulated inTable 1.
( 1 ) ( 2 ) ( 3 )
constant 1.4991** 0.4669** 1.0118**( 0.3345 ) ( 0.19 ) ( 0.2113 )
minority 0.2877** 0.2659** 0.2763**( 0.0228 ) ( 0.0247 ) ( 0.0173 )
Log 25th %ile value -0.0535** -0.0559** -0.0505**( 0.0102 ) ( 0.0106 ) ( 0.0135 )
Log appraised home value -0.0334* 0.4567** 0.4504**( 0.0151 ) ( 0.0314 ) ( 0.0427 )
Log initial credit limit -0.0916** -0.5372** -0.5549**( 0.0213 ) ( 0.0323 ) ( 0.0424 )
Poverty 0.1203 0.1402* 0.2866**( 0.0747 ) ( 0.0755 ) ( 0.1037 )
Log Zip price/price in 1/2000 0.1108**( 0.0211 )
Adj. R-sq. 0.11 0.11 0.09degrees of freedom 316,287 207,646 129,631Exclude value capped loans? N Y Y
Table 5: Mean Zip Code fraction of black or Hispanic homeowners and median home valueas of 2000 across all HECM originations by yearYear Mean Minority Mean Log 2000 Median Value
2000 0.19 11.902001 0.18 11.932002 0.19 11.912003 0.20 11.922004 0.23 11.892005 0.23 11.862006 0.23 11.812007 0.25 11.722008 0.27 11.732009 0.21 11.882010 0.21 11.872011 0.20 11.88
29
Table 6: Log ratio of Zillow Zip Code home price in January, 2006, to January, 2011 regressedon Zip Code characteristics as of 2000 Census and ratio of HECM originations 2004-2007to estimated eligible homeowning households, 2010. Note a larger value for the dependentvalue means a more severe price crash. All specifications include metropolitan area dummyvariables (which subsume the “Sand State” main effect). Standard errors clustered at themetropolitan level
( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 )
constant 0.0177 0.5722** 0.1474 -0.1913 0.9556* -0.0135( 0.0395 ) ( 0.2174 ) ( 0.1911 ) ( 0.209 ) ( 0.4663 ) ( 0.1861 )
Originations 2004-2007Eligible HH 2010 2.9648** 1.5044** 0.8006* 0.6361 0.4246 0.1956
( 0.446 ) ( 0.5602 ) ( 0.4466 ) ( 0.5385 ) ( 0.68 ) ( 0.4945 )Single Family % 0.0148 0.0053 0.0074 0.0081 -0.0253 0.0152
( 0.0348 ) ( 0.0335 ) ( 0.0323 ) ( 0.02 ) ( 0.0275 ) ( 0.0305 )Log owners 65+, 2000 0.0077 -0.0163* -0.0051 0.0229 0.0478 -0.0021
( 0.0095 ) ( 0.0099 ) ( 0.0101 ) ( 0.0188 ) ( 0.036 ) ( 0.0094 )Log owners 65+, 2010 -0.047** -0.0017 -0.0076 -0.018 0.011 -0.0103
( 0.0119 ) ( 0.0121 ) ( 0.0121 ) ( 0.0203 ) ( 0.0251 ) ( 0.0115 )Log owners 2000 0.0154 0.0269** 0.0224** 0.0274**
( 0.0105 ) ( 0.0097 ) ( 0.0094 ) ( 0.0094 )Log owners 2010 0.0321** -6e-04 -0.0015 -0.0044
( 0.0098 ) ( 0.0099 ) ( 0.0101 ) ( 0.0085 )minority 0.193** 0.1798** 0.6614 0.1169* 0.2071**
( 0.0374 ) ( 0.0464 ) ( 0.4785 ) ( 0.0563 ) ( 0.0447 )Log 25th%ile home value -0.0447* -0.0063 0.02 -0.0748* 0.0011
( 0.0193 ) ( 0.0179 ) ( 0.0167 ) ( 0.0417 ) ( 0.0179 )Sand State * minority 0.0873 0.0243 -0.0388 0.0476
( 0.0645 ) ( 0.0397 ) ( 0.0697 ) ( 0.0647 )Sand State * log 25th%ile -0.1542** -0.0028 -0.0102 -0.1279**
( 0.0301 ) ( 0.0128 ) ( 0.0198 ) ( 0.0294 )Log Zillow Zip price 2006/2002 0.3877**
( 0.0454 )Adj. R-sq. 0.84 0.85 0.86 0.83 0.86 0.88degrees of freedom 6537 6535 6533 1186 1187 6532Minority subsample No No No Low High No
30
Related Documents
