Medicaid and the Housing and Asset Decisions of the Elderly: Evidence from Estate Recovery Programs

 

Nadia Greenhalgh-Stanley

Department of Economics and Center for Policy Research Maxwell School of Citizenship and Public Affairs

Syracuse University 426 Eggers Hall

Syracuse, NY 13244-1020 Mobile: (216)832-3554

Fax: (315)443-1081 nngreenh@maxwell.syr.edu 

 

Abstract I examine the impact of Medicaid on elderly housing and portfolio decisions by using recent state-by-calendar-year level variation in the Medicaid treatment of owner-occupied housing assets from the adoption of Medicaid estate recovery programs. Prior to the adoption of these programs, the house, which represents the most important non-pension asset to the elderly, was exempt from determining Medicaid eligibility and served as both a place of residence and a store of wealth. Adoption of estate recovery programs changed the owner-occupied housing safety net by making the house eligible for recovery by the government, which increased the implicit tax of holding owner-occupied housing. Using data from 1993-2004 in the Health and Retirement Study on elderly individuals, I find that state adoption of estate recovery programs makes the elderly decrease homeownership at death by 20 percentage points off a base homeownership rate of 60%, making them 33% less likely to own their homes at death and has a small impact on homeownership rates while the recipients are alive. Also, there is evidence that trusts are treated as a substitute to housing in order to preserve assets and carry out bequest motives at death. Adoption of these programs decreased the housing share of the elderly wealth portfolio.

I would like to thank Gary Engelhardt, Jeff Kubik, Chris Rohlfs and seminar participants at Syracuse University for helpful comments. The research was supported by the Center for Retirement Research at Boston College funded through a grant from the U.S. Social Security Administration as part of the Dissertation Fellowship Program. The opinions and conclusions are solely those of the author and should not be construed as representing the opinions or policy of the Social Security Administration, the Center for Retirement Research at Boston College, The Center for Policy Research or Syracuse University. All errors are my own.

2  

I. Introduction

Medicaid is by far the most important provider of long-term care insurance in the United

States. A key feature of Medicaid, which is a federally mandated program, is that eligibility is

means-tested.1 Specifically, to be eligible, unmarried elderly individuals must have assets and

income below their state’s eligibility levels. Persons with assets greater than this limit must pay

for nursing home care out of pocket and, hence, spend down their wealth. However, owner-

occupied housing (and primary vehicle) assets are exempt from these rules. In addition,

Medicaid traditionally did not go after housing assets when a recipient died, making housing a

particularly attractive asset to hold for those elderly with a bequest motive. This has led to a

great deal of interest among economists and policy makers on the extent to which Medicaid

distorts the housing, portfolio, and bequest decisions of the elderly by creating a subsidy for

holding owner-occupied housing. This is particularly important because owner-occupied

housing is the largest non-pension asset for the elderly.

Unfortunately, there is relatively little empirical evidence from the public and urban

economics literatures on the impact of Medicaid means-testing on elderly housing behavior.

Moreover, a fundamental problem with the few studies that exist is that the estimates of

Medicaid’s impact on housing have relied on cross-state variation in policies and might have

been confounded by other factors that vary across states and also affect elderly housing behavior.

In this paper, I attempt to circumvent some of the difficulties that have plagued previous

studies on the impact of Medicaid on elderly housing and wealth decisions by using recent state-

by-calendar-year level variation in the Medicaid treatment of owner-occupied housing assets                                                             1 Medicaid is social insurance designed to provide health insurance and long-term care for the poor, which should not be confused with Medicare, which provides acute care for the elderly. While Medicaid is a federally mandated program, besides being subject to a few broad mandates, states have some lee-way in their state programs.

3  

from the adoption of Medicaid estate recovery programs. Under these programs, states have the

right to reclaim the value of Medicaid expenditures on nursing home care after the death of the

individual by placing liens on the homes of Medicaid beneficiaries. In particular, the Omnibus

Budget Reconciliation Act of 1993 (OBRA93) mandated that all states adopt estate recovery

programs to recoup assets from the estates of Medicaid recipients. This law had important

implications for the housing assets: prior to the adoption of estate recovery programs, the house

was deemed a safe asset; after the adoption, it was subject to recovery by the government to

repay states for benefits received, once the recipient was deemed permanently institutionalized.

The adoption of an estate recovery program makes housing a less attractive asset in the

portfolio of the elderly for two reasons. First, while housing assets are still exempt from the

Medicaid eligibility decision, the state now can recover the value of Medicaid expenditures by

going after the house, most often with a lien, resulting in bequests being less likely to be made

through the house. Second, homeownership (and the holding of home equity) in the portfolio

are less attractive overall, as the implicit tax on housing assets under Medicaid has risen.

Overall, if the elderly are responsive to Medicaid rules in their housing, portfolio, and wealth

decisions, then the timing of the adoption of estate recovery programs across states should have

affected the holdings of home equity and the frequency of homeownership and housing-related

bequests among the elderly.

While twenty-six states had estate recovery programs prior to 1993, most of the

remaining states began to adopt such programs after OBRA93. In fact, by 2004, forty-seven

states had estate recovery programs. However, because states had a vast amount of control over

the type and size of their estate recovery programs, there has been wide variability in the

structure of programs across states. By far, the most popular component is known as a TEFRA

4  

lien, which is placed on the home while the owner is living, enabled in the Tax Equity and Fiscal

Responsibility Act (TEFRA) of 1982. In 1993, five states allowed for TEFRA liens; by 2004,

nineteen allowed for such liens. Overall, because different states adopted both their estate

recovery programs and TEFRA lien features in different calendar years, respectively, there is

substantial state-by-time variation in Medicaid’s treatment of owner-occupied housing assets,

since 1993.

I use detailed panel data on elderly housing, portfolio, and bequest behavior from the

Health and Retirement Study (HRS), to provide new evidence on the impact of Medicaid and

means-testing on asset and portfolio decisions and to perform, what is to the best of my

knowledge, the first empirical evidence on the impact of estate recovery programs on elderly

behavior. I also capitalize on the so-called “exit” interviews done by the HRS, which consist of

information gathered on the value and asset composition of estates and bequests from next of kin

after an HRS respondent dies (“exits”). These unique and previously unused data provide

detailed information on end-of-life decisions.

There are three primary findings. First, state adoption of ERPs (or TEFRA liens) induces

the elderly to decrease their homeownership at death by 20 percentage points. On a base

homeownership rate of 60%, this represents a 33% decrease in the homeownership rate

(20/60=0.33) when their state adopts a TEFRA lien or ERP. Second, I find suggestive evidence

that trusts are used as substitutes to housing to carry out bequest motives, as state adoption of

ERP or TEFRA liens results in an increase of trust participation of 9 percentage points. On a

base trust participation rate of 16%, this represents a 58% increase in trust participation at death.

Finally, I find that adoption of these programs results in a decrease in the proportion of the total

5  

wealth portfolio that is made up of primary housing assets, which is consistent with the past

literature.

The paper is organized as follows: Section II describes the past literature and

institutional details of Medicaid and its’ associated estate recovery program. Section III

discusses the data construction and empirical framework. Sections IV, V, and VI give the

estimation results for housing, trust, and portfolio decisions, respectively. There is a brief

conclusion.

II. Background

In 2000, the typical elderly person faced an average cost of $50,000 annually for nursing

home care. Dick et al. (1994) estimate that 12% of people needing a nursing home will require

its services for more than five years. Generally speaking, these costs are considerable and can be

unexpected towards the end of life. When facing these large and uncertain expenses of a nursing

home and long-term care, which are the type of expenses older individuals should insure

themselves against, the elderly have four options for insurance. First, they can self-insure by

accumulating assets as a buffer for these uncertain future nursing home costs. Secondly, they

can self-insure by having their kids provide informal long-term care services, which can be

carried out through either intergenerational or inter vivos transfers, in which care is provided in

exchange for bequests. Third, they could buy market services in the form of private long-term

care insurance. However, the private long-term care insurance market has been historically

small due to supply- and demand-side failures (Brown and Finkelstein, 2007). This market also

suffers from adverse selection problems, meaning insurance companies fear that only the really

6  

sick will buy the policies, making it more expensive for them to provide insurance. Finally, they

can use government provision of insurance.

Once the government intervenes in the provision of long-term care insurance, crowd-out

of private behavior becomes an issue. The extent to which social insurance crowds out private

behavior is a perennial topic in economics. In this situation, government crowd-out of private

behavior can happen in two ways. First, after the availability of government insurance,

individuals who previously had chosen to self-insure through the accumulation of assets would

anticipate the government coverage and decide to accumulate fewer assets. In other words, there

would be an implicit spend-down of assets due to the expectation of government financed

insurance, which could be carried out through higher consumption in the years prior to a nursing

home. Second, Medicaid’s role as a payer of last resort may directly crowd out demand for

private long-term care insurance, see Brown and Finkelstein (2008). To better understand how

spend-down works, I next lay our basic rules for eligibility.

Medicaid Eligibility and Spend-Down Laws

The primary way the elderly qualify for Medicaid payment of nursing-home costs is

through the Medically Needy Programs.2 These programs allowed states to expand the elderly

population beyond those eligible to qualify for cash-assistance programs, i.e., the really poor

elderly. Medically Needy programs were in place in 29 states in 2003. They are often

considered the most generous because they allow individuals with higher incomes to qualify for

Medicaid on the condition that their current income is inadequate to pay for their medical

expenses. This allows those with much higher incomes to be eligible for Medicaid by spending

                                                            2 There are five ways in which an elderly individual can qualify for Medicaid (Congressional Research Service 1993), see Appendix A for a full description.

7  

down their assets and income on medical care in order to be deemed needy. The Medically

Needy elderly are those who policy makers and economists focus their attention on when

considering spend-down and its associated consequences, because these are the elderly with

enough assets for the spend-down thresholds to be binding.

More specifically, Medicaid spend-down laws are a complicated set of rules that are

comprised of different asset and income regulations by state. There are two parts to spend down

for Medicaid eligibility: an individual must meet asset eligibility tests as well as income

eligibility tests. Spend-down of assets occurs as individuals are required to contribute liquid

resources toward the cost of their care until the typical state threshold of $2000 is reached.3

Importantly, this limit excludes the home and primary vehicle.

Spend-down of income occurs as an individuals’ excess income, defined as income above

the state’s income threshold, is placed toward the cost of their care. The income spend-down is

more complicated because of the different ways in which elderly qualify for Medicaid. In order

to ensure that individuals do not simply gift their homes to relatives or sell them immediately

prior to entering a nursing home, there is a thirty-six month look back period, in which all

homes/assets gifted or sold within three years of applying for Medicaid are still eligible toward

Medicaid eligibility and recovery.4

Medicaid imposes a 100% implicit tax on holding financial assets above $2000.

Simultaneously, Medicaid exempts owner-occupied housing assets from the Medicaid eligibility

decision. As a result of the differential treatment of assets, Medicaid generates a lower implicit

tax on holding owner-occupied housing assets, which gives an incentive for the elderly to hold

                                                            3 The $2000 limit is adjusted for married couples. 4 In 2006 it was extended to five years.

8  

onto housing assets. Estate recovery programs changed the implicit tax of holding owner-

occupied housing associated with Medicaid’s differential treatment of assets. These programs

are described in detail below.

Estate Recovery Programs

Due to the large share of public expenditures devoted to Medicaid in the 1980s, the

federal government passed OBRA93, which mandated that all states adopt estate recovery

programs (ERPs) to recoup assets from the estates of Medicaid recipients.5 OBRA93 laid out

that the minimum that could be recovered from an estate was the probate estate, and the

maximum that could be recovered was the amount left in an estate.6 Other than a few structured

mandates set up by the federal government, the states had a vast amount of control over the type

and size of their estate recovery program, which led to wide variability of programs across states.

This law had important implications for the house, which prior to 1993 was deemed a safe asset,

but was now subject to recovery by the government to repay states for benefits received once the

recipient was deemed permanently institutionalized. It also had implications for wealth holdings

among the elderly because their largest non-pension asset was now subject to recovery.

Existing Studies

There are four strands of existing literature that are particularly relevant.7 First, Brown

and Finkelstein (2008) and Brown, Coe, and Finkelstein (2007) examine Medicaid’s crowd-out

                                                            5 In 2002 the government paid out $31 billion in Medicaid long-term care costs, which according to the Kaiser foundation made up 52% of Medicaid spending and only provided services to 7% of Medicaid recipients. As a result, the federal government decided to adopt cost containing measures to help states balance their budgets. 6 A probate estate consists of any assets that can be left in a will. 7 While there is an extensive literature on the health aspects of long-term care (e.g. Norton, 2000) and the impact of social insurance on labor supply (e.g. Krueger and Meyer, 2002), there is virtually no empirical evidence on the affect of social insurance on housing decisions among the elderly. Dietz and Haurin (2003) have a comprehensive

9  

of demand for private long-term care insurance. They generally concluded that Medicaid results

in crowd-out of demand for private long-term care insurance, mainly due to Medicaid’s role as

the payer of last resort.

The second consists of studies investigating the impact of other social insurance

programs on the housing and living arrangements of the elderly. The basic finding of the

literature has been that elderly living arrangements are responsive to Social Security benefits.8

In particular, Engelhardt, Gruber, and Perry (2005) and Engelhardt (2008) both find an impact of

Social Security on the headship decision as well as the homeownership decision. Engelhardt and

Greenhalgh-Stanley (2008) also provide evidence that the homeownership and shared living

arrangement decisions are affected by changes in the reimbursement generosity of Medicare

home health benefits. These studies provide evidence that social insurance benefits do, in fact,

affect elderly housing decisions.

The third deals with a set of studies that investigate Medicaid’s spend-down feature on

wealth accumulation of the elderly. Using variation in measures of state Medicaid generosity,

Coe (2007) found that unmarried respondents decreased total wealth, while married couples

seemed to transfer assets from unprotected financial assets to protected housing assets, through

some parts of her wealth distribution. 9

III. Econometric Framework, Identification, and Data Construction

                                                                                                                                                                                                survey of the recent literature in housing economics, but only 11 of 251 articles were related to elderly housing research. 8 See Michael, Fuchs, and Scott (1980), McGarry and Schoeni (2000), Engelhardt, Gruber, and Perry (2005), and Costa (1999). 9 Finally, there is a strand of literature involving estate taxation and planning. This is relevant due to the fact that non-housing assets are essentially taxed away under the Medicaid spend-down rules. At the end of life, the elderly have tried to convert assets into types that receive favorable treatment with estate taxes. Poterba (2001) provides an overview of this literature.

10  

The estimates of the past literature, in the end, rely on cross-state variation and might be

confounded by other factors that vary across states that also affect elderly housing behavior. I

overcome the problems that have plagued past literature by exploiting state-by-time variation in

state adoption of estate recovery programs (ERPs) and TEFRA liens. TEFRA liens allow a lien

to be placed on the home of a permanently institutionalized recipient, which serves to keep the

states’ vested interest in the estate and track the use of the house.

TEFRA liens have the unique nature of being the only type of estate recovery program to

be enacted while the Medicaid recipient is still alive, which may affect behavior differently than

ERPs in general. ERPs may be interpreted as “far off” because they are implemented after the

death of a recipient. Because TEFRA liens are placed on the house while the recipient is still

alive, they may trigger estate planning and more immediate behavioral responses. I study the

impact of state adoption of ERPs and TEFRA liens on housing and asset portfolio decisions

among the elderly. Furthermore, ERPs and TEFRA liens provide different incentives for

Medicaid recipients by marital status because by law they cannot be enacted on a house when

there is a surviving spouse.

Impact of Recovery on the Unmarried Elderly

To better understand how estate recovery programs work, consider an unmarried elderly

individual living in a state without a recovery program. To become eligible for Medicaid

payment for nursing-home services, the respondent must spend down her assets to meet her state

asset threshold and delegate extra income to paying for her care. However, the house is not

subject to the asset threshold and can be held until death. An important feature is that the house

can jointly serve as a residence and a store of wealth that then can be bequeathed. At death, the

11  

house is bequeathed to a delegated relative or charity, which is usually set out in a will. In

contrast, under an estate recovery program, the same Medicaid asset and income eligibility

requirements apply, and the value of the house is not used toward determining Medicaid

eligibility, but at death, the house is now eligible for recovery by the state and cannot be

bequeathed to a source of choice.

ERPs and Medicaid provide an incentive for unmarried recipients to sell their homes,

since they are not protected from recovery and will be used to help pay for their care. They may

also provide incentives to decrease their home equity due to the fact that the government will be

getting the value of the home. This could be carried out through a decrease in the share of

owner-occupied housing assets in the wealth portfolio.10

Impact of Recovery on the Married Elderly

Married individuals, however, do not face any incentive to change homeownership while

the recipient is still alive. They do face incentives to retain homeownership at death, though.

Prior to recovery, married individuals were able to dispose of their home in any manner they saw

fit, i.e. bequeath their home to a relative or charity. Post recovery, married individuals were only

able to pass on their homes if they retained ownership at death, otherwise the government would

have been able to recover the value of the home after death. Similarly if a married individual

lived in a non-TEFRA state and non-ERP state, they would be able to bequeath their house to

any source of choice even after receiving Medicaid payment for long-term-care services. After

recovery, the house can only remain a store of wealth by retaining homeownership with their

                                                            10 One way to do this would be to decrease home equity through lack of maintenance and upkeep as explained in Davidoff (2004) and is consistent with the findings of Coe (2007) when she measured the impact of Medicaid generosity by state on household savings and portfolio decisions.

12  

spouse. As a result, there is a strong incentive for the married sample in ERP and TEFRA states

to retain homeownership at death.

Researchers may believe that married individuals begin to act like unmarried individuals

(with regard to ERPs) once their spouse moves into a nursing home. However, in practice, the

remaining married spouse (community spouse) in the home is ignored by the state because it is

too expensive to track the home for the duration of the community spouses stay in the house.

Instead, the house in practice is only recovered against if the community spouse later needs

nursing home care provided by Medicaid. In that case, the community spouse would have to

spend-down their wealth and income to be eligible for Medicaid according to their states

unmarried thresholds and then the house would be eligible for recovery after care was provided.

However, if the community spouse never needed Medicaid coverage of long-term care services,

in practice the home would not be recovered against. Hence, rational married community

spouses would not need to act as if they were unmarried in regards to ERPs.

Regression Specification and Data Construction

To examine the relationship between state adoption of ERPs and TEFRA liens and

behavioral outcomes empirically, let i, s, and t index household, state, and calendar-year

respectively. Then the homeownership-at-death regression specification is

(1) * *

,

Outcome ERP Unmarried TEFRA Unmarried ERP TEFRA Unmarriedist ist ist st st i

ist s t ist

D D D D D Du

β δ η λ σθ γ φ

= + + + +

+ + + +X

where OutcomeD is a dummy variable that takes on a value of one if the respondent owns their

home at death and a zero otherwise, ERPD is a dummy equal to one if state s has an ERP in

calendar-year t, TEFRAD takes on a value of one if state s has a TEFRA lien in place in calendar-

13  

year t, UnmarriedD takes on a value of one if respondent i is unmarried, γ is a full set of state fixed

effects, φ is a full set of calendar-year fixed effects, and u is a disturbance term. In addition, X

includes dummy variables for gender, non-white, college degree or higher, the number of

independent activities of daily living (IADLs), as well as age, and number of children.11 The

explanatory variable of note, *ERP UnmarriedD , takes on a value of one if an unmarried respondent

lives in a state s with an estate recovery program (ERP) in calendar-year t and a zero otherwise.

The coefficient of interest, β , represents the impact for unmarried respondents relative to

married respondents in ERP states relative to non-ERP states on dying while still owning a

home. The other explanatory variable of note, *TEFRA UnmarriedD , takes on a value of one if the

unmarried respondent lives in state s that has a TEFRA lien program in calendar-year t and a

zero otherwise. The other coefficient of interest, δ , represents the impact for unmarried

respondents relative to married respondents in TEFRA states relative to non-TEFRA states on

dying while still owning a home. The total impact of an estate recovery program on

homeownership at death is represented by the coefficient, β δ+ .

If estate recovery programs and TEFRA liens do have an impact on homeownership

decisions among the elderly, we would expect to see them for the unmarried sample that faces

the incentives to sell their homes at death. As a result, I would expect ,β ,δ and β δ+ to be

negative coefficients depicting a negative relationship between state adoption of these programs

and homeownership rates among the unmarried relative to married elderly individuals. These

expected negative coefficients would reflect both the unmarried incentive to sell their home at

death and the married samples incentive to retain homeownership at death.

                                                            11 IADLs are based on a scale of zero to five independent daily living activities the recipient can do on their own.

14  

I estimate the parameters in equation (1) using the AHEAD cohort of the Health and

Retirement Study (HRS). The HRS is a pre-eminent data set on economics and demographics on

aging. It collects a wealth of information on family structure, wealth portfolios, and

demographics for a nationally representative sample. The AHEAD cohort consists of elderly

individuals seventy and older in 1993. They are surveyed every two years. I create a panel data

set that consists of AHEAD homeowners in 1993 and follows them until 2004 or their death.

Because I am relying on state-by-time variation for identification, I need to use the restricted-

access HRS data to attach the state of residence to each recipient. I collected data on state

adoption of TEFRA liens and ERPs due to the lack of existing data in the literature. Appendix B

gives a full description of data collection. Sample descriptive statistics are presented in Table 1.

IV. Estimation Results

The primary outcome of interest is homeownership at death. In order to estimate the

impact of state adoption of ERPs and TEFRA liens on homeownership at death, I estimate

Equation (1) by exploiting the exit data of the HRS and using linear probability estimation. First,

I begin by isolating the impact of state adoption of only an ERP on homeownership at death by

estimating Equation (1) without the TEFRA lien variables. The results are shown in Column (1)

of Table 2 with standard errors clustered by both state and person identifiers using the Cameron,

Gelbach, and Miller (2006) multi-dimension clustering technique. I cluster by person because

for the same person there is potential autocorrelation over time for the home ownership decision.

I cluster by state because of the worry that there is autocorrelation within the same state over

time. Clustering adjusts for the possibility that observations are not independent. State adoption

of ERPs makes the unmarried relative to the married sample decrease their homeownership at

15  

death by 15.5 percentage points.12 On a base homeownership rate of 60% this represents a 26%

decrease in the homeownership rate at death (15.5/60=0.26).13

While ERPs have a negative impact on dying while retaining home ownership status, it is

possible that the unique nature of TEFRA liens - one type of estate recovery program, and the

only type to be placed on the house while the recipient is still alive – has a differential impact

from general ERPs on elderly behavior. Column (2) of Table 2 shows the results when TEFRA

liens are added back into the estimation. I find that state adoption of ERPs makes the unmarried

sample relative to the married sample decrease homeownership by 13 percentage points. On a

base homeownership rate of 60%, this results in a 22% decrease in the homeownership rate.

State adoption of TEFRA liens makes the elderly decrease homeownership by 6.5 percentage

points. On a base homeownership rate of 60%, this results in an 11% decrease in the

homeownership rate.

While the coefficient for TEFRA liens is not significant, it is appears that the unique

dimension of TEFRA liens being placed on the home while the recipient is still alive does impact

elderly behavior. The total impact of recovery on the unmarried elderly is the sum of the

TEFRA lien coefficient and the ERP coefficients. As a result, the total impact of recovery is a

decrease in homeownership by 19.6 percentage points. On a base homeownership rate of 60%,

this results in a 32.5% decrease in the homeownership rate. Column (3) adds a basic set of

demographic control variables to the analysis. State adoption of TEFRA liens makes the

unmarried elderly relative to the married elderly decrease homeownership by 7.6 percentage

points, while state adoption of ERPs makes the unmarried elderly relative to the married elderly

                                                            12 Naturally more populated states appear in the HRS with more frequency, for subsets of states with sufficient cell size, probit and linear probability estimates are qualitatively similar. 13 I divide the percentage point change by the sample mean to calculate the percent change (marginal effect). 

16  

decrease homeownership by 12 percentage points. The total impact is a decrease in

homeownership by 19.5 percentage points. On a base homeownership rate of 60%, this results in

a 32.5% decrease in the homeownership rate at death.

However, it is possible that other factors also impact the homeownership rate at death

that are not included in the analysis, which would result in omitted variable bias. Consequences

of omitted variable bias include biased and inconsistent parameter estimates. One potential

omitted factor is local housing prices, which may influence the housing tenure decision among

the elderly. As a result, Column (4) includes a housing price index into the estimation to test for

omitted variable bias. Data on a housing price index (HPI) were collected from the Office of

Federal Housing Enterprise Oversight. Column (4) shows that state adoption of TEFRA liens

makes the unmarried elderly relative to the married elderly decrease homeownership by 7.7

percentage points, while state adoption of ERPs results in a decrease in homeownership by 11.8

percentage points. The total impact of recovery is a decrease in homeownership of 19.5

percentage points. On a base homeownership rate of 60%, this results in a decrease in the

homeownership rate at death of 32.5%. The estimates on the impact of recovery on

homeownership rates at death are robust to including the housing price index into the estimation,

meaning that the results in Column (3) do not suffer from an omitted variable bias attributable to

local housing prices.

Another potential problem with the analysis is that when states adopt TEFRA liens and

ERPs, they are also changing other parts of the Medicaid program. They could be changing the

overall Medicaid generosity at the same time as adopting these recovery programs. This would

be another potential source of omitted variable bias of the initial results. I measure the overall

generosity of Medicaid by collecting data on the state Medicaid expenditures over time from

17  

Medicaid Statistics, Programs and Financial Statistics 1993 and Bureau of the Census and data

from the Kaiser Commission from www.cms.hhs.gov/NationalHealthExpendData. Column (5)

tests to see if state Medicaid expenditure is an omitted variable in the initial analysis. State

adoption of TEFRA liens makes the unmarried elderly relative to the married elderly decrease

homeownership at death by 8 percentage points, while state adoption of ERPs results in a

decrease in homeownership at death of 11.9 percentage points. The overall impact is a decrease

in homeownership at death of 19.9 percentage points. On a base homeownership rate at death of

60%, this results in a decrease in the homeownership rate at death of 33.1%. The results are also

robust to adding this variable to the specification.

Finally, trends over time in states may be correlated with elderly portfolio behavior and

other laws passed in these states. There could be a worry that these linear state-by-time trends

are an omitted variable resulting in biased and inconsistent parameter estimates. Column (6)

augments the previous estimation by adding in linear state-by-time trends to the estimation in

order to test for another potential omitted variable bias. I would expect adding these trends to

soak up variation since my estimation relies on state-by-time variation and that the standard

errors could get larger. State adoption of TEFRA liens results in the unmarried elderly relative

to the married elderly to decrease homeownership at death by 7.9 percentage points, while state

adoption of ERPs decreases homeownership at death by 12.6 percentage points. The total impact

of recovery is a decrease in homeownership at death of 20.6 percentage points. On a base

homeownership rate of 60%, this results in a decrease in homeownership at death of 34.4%.

After numerous robustness checks, it is clear that there is a strong negative relationship between

the homeownership rate at death and state adoption of ERPs and TEFRA liens.

18  

This analysis presents strong evidence that the unmarried sample responds to the

incentives set up by the laws and does significantly decrease home ownership rates at death

when faced with an ERP or TEFRA liens program. It is important to note that this is a net effect

of the unmarried sample relative to the married sample. The married sample face strong

incentives at death to retain homeownership. Prior to adoption of ERPs and TEFRA liens, they

could protect the home for relatives whether they retained ownership at death or not, but after

ERPs and TEFRA liens were adopted, married respondents could only protect the home for

relatives if they died owning it.

V. Impact on Trusts at Death

Because owner-occupied housing assets can no longer be used as a store of wealth to

carry out bequest motives for the unmarried elderly, they could look for substitute methods to

pass on assets, such as trusts. The unmarried elderly face an incentive to die with assets in a trust

in order to protect assets from recovery. Due to the fact that married respondents have the home

as a guaranteed way to protect assets for their heirs and satisfy their bequest motives, they do not

face the same incentive to put assets in a trust. In order to test whether the unmarried elderly use

trusts as a substitute asset protection vehicle, I estimate equation (1) using a dummy variable for

whether the respondent died with assets in a trust as the dependent variable. The results are

shown in Table 3, which is formatted in the same manner as Table 2.

Column (1) shows the estimation when only ERPs are included in the estimation and

there are no controls. State adoption of ERPs makes the unmarried relative to the married

sample increase their trust participation rate by 4.1 percentage points. On a base trust

participation rate of 15.7% at death, this results in the elderly being 26% more likely to die with

19  

assets in a trust. Column (2) presents the results when TEFRA liens are added to the estimation.

State adoption of TEFRA liens make the unmarried elderly relative to the married elderly

increase trust participation at death by 8.5 percentage points. On a base trust participation rate at

death of 15.7%, this results in a 54% increase in the trust participation rate at death. State

adoption of ERPs make the unmarried elderly relative to the married elderly increase trust

participation at death by 1 percentage point. On a base trust participation of 15.7%, this results

in a 9.6% increase in trust participation. The total impact of recovery, the sum of TEFRA liens

and ERP coefficients, is an increase in trust participation at death of 9.5 percentage points or

60.5%.

Column (3) adds in a basic set of demographic controls to the estimation. State adoption

of TEFRA liens results in the unmarried relative to the married elderly increase trust

participation at death by 7.7 percentage points, while state adoption of ERPs make the elderly

increase trust participation at death by 1.5 percentage points. The total impact of recovery is an

increase in trust participation at death by 9.2 percentage points. On a base trust participation rate

at death of 15.7 percent, this results in an increase in the trust participation rate at death of

58.6%.

As with the homeownership at death analysis, there is a concern about potential omitted

variable bias. As a result, Columns (4)-(6) add elements to the specification to test for this.

Column (4) adds a housing price index to the estimation. State adoption of TEFRA liens results

in the unmarried elderly relative to the married elderly increase trust participation at death by 7.3

percentage points, while state adoption of ERPs make the elderly increase trust participation at

death by 1.6 percentage points. The total impact of recovery is an increase of 8.9 percentage

20  

points. On a base trust participation rate of 15.7%, this results in an increase in the trust

participation rate at death of 56.7%.

Column (5) adds in state Medicaid expenditure as a proxy for changes in Medicaid

generosity by states. State adoption of TEFRA liens makes the unmarried elderly relative to the

married elderly increase trust participation at death by 7.3 percentage points, and state adoption

of ERPs make the elderly increase it by 1.5 percentage points. The total impact of recovery is an

increase in trust participation at death of 8.8 percentage points or an increase in the participation

rate of 56%.

Finally, Column (6) includes a linear state-by-time trend to the estimation. State

adoption of TERA liens makes the unmarried elderly relative to the married elderly increase trust

participation by 5.8 percentage points and state adoption of ERPs makes the elderly increase it

by 1.3 percentage points. The total affect of recovery is an increase of 7.1 percentage points or

an increase in the trust participation rate of 45.2%. The trust estimates suggest that the

unmarried elderly relative to the married elderly respond more strongly to TEFRA liens, possibly

due to the fact that they are placed on the home while the recipient is still alive, as opposed to

ERPs, which are enacted on the estate and home after the recipient has died. The main result of

the unmarried elderly relative to the married elderly increase their trust participation rate at death

by 58.6% is robust to including different factors in the specification. Because the unmarried

sample does not have the house as a guaranteed asset protection for heirs, they seem to treat

trusts as a substitute asset protector for their heirs and remaining relatives.

VI. Extensions

21  

As the previous results in this paper have shown, state adoption of TEFRA liens and

ERPs clearly alter elderly asset behavior at death. However, it is also interesting to estimate the

affect of these programs on elderly behavior while the recipients are still alive and on different

outcomes.

Impact of Aging

The unique nature of having data on the AHEAD cohort containing both end-of-life

decisions and asset composition while the recipients are still alive, allow me to explore possible

trends in homeownership rates. More specifically, I can estimate the impact of TEFRA liens and

ERPs on the elderly while they are alive, 0-2 years prior to death, and 2-4 years prior to death.

While the recipients are alive, the unmarried elderly have an incentive to decrease owner-

occupied housing assets in their portfolio due to the higher implicit tax of holding owner-

occupied housing placed on the elderly by recovery programs. The elderly can carry this out by

decreasing homeownership or by decreasing their home equity. The married elderly do not have

an incentive to change homeownership while they are alive because recovery cannot be

performed on their estates while they have a surviving spouse. The married elderly provide a

control group in the more traditional sense because they do not face any incentives to change

their homeownership rates while they are alive.

Homeownership Rates

To examine any trends in the impact of estate recovery programs on homeownership, I

use equation (1) with homeownership data while the sample is still alive, data 0-2 years prior to

death, and data 2-4 years prior to death. Table 4 shows the time trends analysis for

homeownership rates for an unbalanced panel. Column (1) shows the results from Column (3) of

22  

Table 2. At death, state adoption of TEFRA liens makes the unmarried elderly relative to the

married elderly decrease homeownership rates at death by 7.6 percentage points, while state

adoption of ERPs results in the elderly decreasing homeownership at death by 11.9 percentage

points. The total impact of recovery is a decrease in homeownership at death of 19.5 percentage

points. On a base homeownership rate at death of 60%, this results in a decrease of the

homeownership rate at death of 32.5%.

Column (2) portrays the results for homeownership with data provided in the interview

prior to death or 0-2 years before death. State adoption of TEFRA liens makes the unmarried

elderly relative to the married elderly decrease homeownership by 1.1 percentage points, while

state adoption of ERPs make the elderly decrease homeownership by 2.9 percentage points. The

total impact of recovery is a decrease in homeownership 0-2 years prior to death by 4.0

percentage points. On a base homeownership rate of 75%, this results in a decrease in the

homeownership rate of 5.4%.

Column (3) shows the affect of these laws on homeownership as reported two interviews

prior to death, or 2-4 years prior to death. State adoption of TEFRA liens make the unmarried

elderly relative to the married elderly decrease homeownership 2-4 years prior to death by 1.5

percentage points, while state adoption of ERPs make the elderly decrease homeownership by

2.6 percentage points. The total impact of recovery is a decrease in homeownership as reported

two interviews prior to death by 4.1 percentage points. On a base homeownership rate 2-4 years

prior to death of 84.1%, this results in a decrease in the homeownership rate by 4.9%.

Column (4) provides the results for the elderly while they are alive without any

constraints put on the sample with regards to timing of death. State adoption of TEFRA liens

23  

make the unmarried sample relative to the married sample decrease homeownership by 2.4

percentage points, while state adoption of ERPs make the elderly decrease homeownership by

1.6 percentage points. The total impact of recovery is a decrease in homeownership by 4.0

percentage points. On a base homeownership rate while alive of 89.4%, this results in a decrease

in the homeownership rate by 4.5%. The negative relationship between recovery and

homeownership becomes more negative as the recipient gets closer to death, with most of the

homeownership changes happening at death.

However, it is possible that the changes seen in homeownership rates across time are

partially attributable to the changing samples over time rather than entirely to the changes in

state adoption of recovery programs. Table 5 shows the time trend results for homeownership

over time with a balanced panel sample. However, these results may suffer from a selection bias

due to the fact that I must restrict this sample to have been alive for 4 years prior to death, or in

other words I restrict my sample to those that were 74 and older when they died.

Column (1) of Table 5 shows the results for this restricted sample at death. State

adoption of TEFRA liens make the unmarried elderly relative to the married elderly decrease

homeownership by 2.6 percentage points at death, while state adoption of ERPs make the elderly

decrease homeownership at death by 2.7 percentage points. The total impact of recovery is a

decrease in homeownership at death by 5.3 percentage points. On a base homeownership rate at

death of 62.2%, this results in a decrease in the homeownership rate of 8.5%.

Column (2) shows the results for homeownership as reported in the interview prior to

death, 0-2 years prior to death. State adoption of TEFRA liens make the unmarried relative to

the married elderly decrease homeownership by 0.4 percentage points, while state adoption of

24  

ERPs make the elderly decrease homeownership by 3.2 percentage points. The total impact of

recovery is a decrease in homeownership by 3.6 percentage points. On a base homeownership

rate of 75.1%, this results in a decrease in the homeownership rate by 4.8%.

Column (3) provides the results two interviews prior to death, 2-4 years prior to death.

State adoption of TEFRA liens make the unmarried elderly relative to the married elderly

decrease homeownership by 1.5 percentage points, while state adoption of ERPs make the

elderly decrease homeownership by 2.6 percentage points. The total impact of recovery is

measured by a decrease in homeownership by 4.1 percentage points. On a base homeownership

rate of 84.1%, this results in a decrease in the homeownership rate by 4.9%. As with the

unbalanced panel estimation results, the behavior change is clearly happening right before death.

Home Equity

While estate recovery programs change the implicit tax on housing assets and therefore

create an incentive to decrease housing, this can be carried out on alternative housing margins,

rather than simply through the homeownership margin. As Davidoff (2004) suggested, the

elderly may keep owning their home but stop performing maintenance on their house, and as a

result decrease their home equity. Table 6 provides the results when home equity (as reported

when alive) is used as the dependent variable. This table is presented in the same format as

Table 2.

Column (1) shows the results without any controls and just considering state adoption of

ERPs. State adoption of ERPs makes the unmarried elderly relative to the married elderly

25  

decrease home equity by $3674.14 On a mean home equity of $119,124, this results in a decrease

of 3.1%. Column (2) considers when TEFRA liens are added to the estimation. State adoption

of TEFRA liens makes the unmarried sample relative to the married sample decrease home

equity by $18,559, while state adoption of ERPs make the elderly decrease home equity by $360.

The total impact of recovery is a decrease of almost $19,000.

Column (3) adds in a basic set of demographic controls. Housing equity is decreased by

the unmarried sample relative to the married sample by $24,813 when their state of residence

adopts TEFRA liens. State adoption of ERPs results in the elderly decreasing home equity by

$2900. Clearly the elderly are responding in a stronger way to TEFRA liens being placed on

their homes while they are still alive. The total impact of recovery is a decrease in home equity

by $27,700. On a base home equity of $119,123, this results in a decrease in home equity by

23.5%.

Column (4) augments the estimation by adding in a housing price index to test whether

local housing prices are explaining part of the negative relationship between recovery programs

and home equity. State adoption of TEFRA liens results in the unmarried elderly relative to the

married elderly decrease home equity by $22,320, while state adoption of ERPs makes the

elderly decrease home equity by $3170. Column (5) includes a proxy for Medicaid generosity.

Housing equity is decreased by the unmarried elderly relative to the married elderly by $22,000

when their state of residence adopts a TEFRA lien, while they decrease it by $3360 when their

state adopts an ERP. Column (6) examines the coefficients when a linear state-by-time trend is

included in the estimation. State adoption of TEFRA liens makes the unmarried elderly relative

                                                            14 It is a well known that for a conditional median estimator small cell size will result in a lack of convergence. Unfortunately, my sample has states with very small cell sizes and my median estimates do not converge.

26  

to the married elderly decrease home equity by $20,407, while state adoption of ERPs makes the

elderly decrease home equity by $2861.

Clearly the elderly are changing their behavior in a stronger way when facing the unique

nature of having a TEFRA lien placed on their home when they are still alive. However, these

results do not distinguish whether this is a total change in home equity or simply a shift in

portfolio assets from home equity to total wealth. As a result, the estimation is extended to

consider the wealth portfolio. In results not shown, the estimation was further broken down to

explore the impact on first, second, and third mortgages, but no significant results or

relationships were found.

Housing Share of Portfolio Wealth

To investigate whether the elderly are changing or shifting portfolio assets, equation (1)

is estimated with the housing share of the total wealth portfolio used as the dependent variable.

If the unmarried are changing their behavior on this margin, we would expect to see a decrease in

housing wealth as a percentage of total wealth because they face incentives to decrease their

home equity now that the government will be getting the value of the home. The results are

shown in Table 7, which is set up in the same manner as Table 2.

Column (1) shows the results for the housing share of portfolio wealth when only state

adoption of ERPs is considered. State adoption of ERPs makes the unmarried elderly relative to

the married elderly decrease their housing share of the wealth portfolio by 3.9 percentage points.

On a base housing portfolio share of 54.9%, this results in a decrease of 7.1%. Column (2) adds

TEFRA liens into the estimation. The housing share of the wealth portfolio is decreased by 8.3

percentage points when a state adopts a TEFRA lien, while state adoption of ERPs make the

27  

unmarried elderly decrease it by 0.8 percentage points. The total impact of recovery is a

decrease in the housing share of the wealth portfolio by 9.1 percentage points. On a base

housing share of portfolio wealth of 54.9%, this results in a decrease of 16.6%.

Column (3) adds a basic set of demographic controls to the estimation. State adoption of

TEFRA liens make the unmarried elderly relative to the married elderly decrease housing share

of the wealth portfolio by 8.1 percentage points, and state adoption of ERPs makes the elderly

decrease it by 0.2 percentage points. Column (4) adds a housing price index to the estimation.

State adoption of TEFRA liens makes the unmarried sample relative to the married sample

decrease housing share of portfolio wealth by 8.0 percentage points, while state adoption of

ERPs makes the elderly decrease it by 0.3 percentage points. The total impact is a decrease of

8.3 percentage points or 15.2%.

Column (5) adds state Medicaid expenditures to the analysis. State adoption of TEFRA

liens makes the unmarried sample relative to the married sample decrease housing share of

portfolio wealth by 8.2 percentage points, while state adoption of ERPs makes the elderly

decrease it by 0.2 percentage points. Column (6) adds in state-by-time linear trends. Housing

share of portfolio wealth is decreased by 8.1 percentage points when TEFRA liens are adopted

and by 0.3 percentage points when ERPs are adopted. The results are robust to including

additional controls. The elderly are clearly responding stronger to the TEFRA liens being placed

on their homes while they are still alive. These results suggest that there is some change in the

portfolio composition of housing from these laws while the recipients are still alive.

VII. Conclusions

28  

There are three primary behavioral changes induced by Medicaid’s higher implicit tax of

holding owner-occupied housing resulting from state adoption of estate recovery programs.

First, state adoption of ERPs or TEFRA liens induces the elderly to decrease their

homeownership at death by 20 percentage points. On a base homeownership rate of 60%, this

represents a 33% decrease in the homeownership rate (20/60=0.33) when their state adopts a

TEFRA lien or ERP. Second, state adoption of ERP or TEFRA liens results in an increase of

trust participation of 9 percentage points. On a base trust participation rate of 16%, this

represents a 58% increase in trust participation at death. This is suggestive evidence that trusts

are considered as substitutes to housing to carry out bequest motives by the elderly. Finally, I

find that adoption of these programs results in a 16.6% decrease in the proportion of the total

wealth portfolio that is made up of primary housing assets.

There are a few potential caveats in this estimation analysis. First, I use a dummy

variable for whether a state has an ERP or TEFRA lien. However, there is some evidence that

there is differential enforcement of these laws by state. I am interested in making an index to

augment my analysis to also consider the impact of state enforcement on asset decisions.

Second, I use data on the AHEAD cohort, but it is possible that different cohorts respond

differentially to these laws. In order to address this, the estimation strategy could be applied to

younger cohorts in the HRS sample. Third, these estimates are from a reduced form empirical

analysis. There is not a structural or theoretical model in the analysis. Future work could lay out

a structural model and run a simulation analysis or measure the deadweight loss to society.

While it is clear that the elderly are responding to the estate recovery programs, it is not

clear what the direct policy implication of this is. Because there is not a structural model in my

analysis, these results do not shed any light on what the optimal policy should be. However, if

29  

the true goal is for states to recoup assets to offset their large Medicaid expenditures, then these

estimates can say something about the impact of structuring the programs in this way. One

margin to consider is a cost-benefit analysis for states. More specifically, how much does it cost

for states to implement these programs versus how much they are actually bringing in from

recovery. Karp, Sabatino, and Wood (2005) provide information that the amount recovered as a

percentage of LTC expenditures ranged from 0.01% (LA) to 1.83% (ID), while the annual

administration costs had a range between $30,000 (OK) to $1,645,868 (CA) per year. Only nine

states reported statistics for the administration costs as a percentage of the amount recovered.

The range was from 1.5% (NC) to 11.77% (TN). However, these programs are relatively young

and inadequate records on the number and amount of recoveries have been kept. Hopefully in

the future, a more in depth cost-benefit analysis can be executed on complete recovery data from

each state.

30  

Appendix A

There are five ways in which an elderly individual can qualify for Medicaid (Congressional Research Service, 1993). The first path to Medicaid eligibility is through Supplemental Security Income (SSI), which is a means-tested social insurance program designed to assist poor (disabled) individuals. To qualify for SSI, individuals must have an income below the SSI threshold, $448 per month for an individual and $726 per month for a couple in 1993, and assets below $2000 for an individual and $3000 for a couple in 1993.15 SSI recipients must be covered by all states, unless the states opt to use more restrictive thresholds that were in place in 1972. This is the second path, and these states are called 209(b) states. There were eleven of them in 2003. Since 209(b) states use more restrictive eligibility standards, they are required to determine eligibility after an individual deducts their medical costs from their current income.

The third path is also determined by specific income limits and is often known as the “300 percent rule”.16 It deems that individuals are eligible for Medicaid if their income is within 300 percent of the SSI income level stated above and was used by 35 states in 2003, though some of these states overlapped with the Medically Needy program discussed below. These states are often considered less generous because an individual may not spend down their income in order to qualify. For example, if the threshold is $1344, 300% of $448 benefit in 1993, and an individual has an income of $1345, the individual does not qualify for Medicaid and must pay for their own long-term care costs (Congressional Research Service, 1993).

The fourth path, Medically Needy programs, allowed states to expand the elderly population beyond those eligible to qualify for cash-assistance programs, i.e., the really poor elderly. Medically Needy programs were in place in 29 states in 2003. They are often considered the most generous because they allow individuals with higher incomes to qualify for Medicaid on the condition that their current income is inadequate to pay for their medical expenses. This allows those with much higher incomes than the SSI levels to be eligible for Medicaid by spending down their assets and income on medical care in order to be deemed needy.

The fifth path to Medicaid eligibility is through Qualified Medicare Beneficiaries (QMB). Individuals qualify through this route by being Medicare enrollees with resources less than double the SSI limits and income below the Federal poverty line. States are required to pay Medicare premiums along with co-pays and are allowed to offer full Medicaid coverage if they choose to. Another source of optional state coverage comes from waivers to pay for home and community care on behalf of individuals that would qualify for Medicaid coverage if they were receiving institutionalized care instead.

                                                            15 The income thresholds are adjusted each year by the CPI index to control for inflation. 16 It is called the 300% rule, even though three states choose to use a lower level than this. In 2003, Missouri used 175%, Delaware used 250%, and New Hampshire used 244%, (Coe, 2007).

31  

Appendix B

Due to the lack of existing data, I compiled a data set of state by time variation on ERP and TEFRA lien adoption. Since states do not always keep detailed records of the history of their estate recovery programs, I decided it was more productive to look into government or interest group commissioned studies. Most of these studies were commissioned by the AARP Public Policy Institute and performed by the ABA and they usually involve contacting the states’ Medicaid programs directly and practitioners in the state familiar with the legal aspects to get as much information as possible. Unfortunately, many states did not keep records so they were not able to provide a lot of the requested information and a few states declined to participate in the studies in various years.

I was able to collect data containing information on which states had ERPs and TEFRA liens in 1993, when mandatory estate recovery OBRA93 was passed into law from the article “Picking the Bones of the Poor” by Schwartz and Sabatino (1993). In this article, there were many case studies and descriptions of certain state practices in an appendix, which were used to decipher which states used TEFRA liens as opposed to other types of liens (i.e. post death, etc). In 1995, a survey by ABA was conducted and included a list of the states with TEFRA liens. In 1998, the longtermcarelink.com website Table 1 was used to document which states had TEFRA liens. In 2004, a follow up to the 1995 survey was commissioned to explore the current state of estate recovery programs and to document the growth in numbers and scope of the recovery programs. An inordinate amount of detail was included in “Medicaid Estate Recovery: A 2004 Survey of State Programs and Practices” by Karp, Sabatino, and Wood (2005), which was very helpful and hopefully will help with future studies by adding in more state variation. Unfortunately, the in depth analysis and details were only included in the 2004 surveys and cannot be included in the current panel analysis. I was able to gather 1993, 1995,1998, and 2004 information on estate recovery programs and TEFRA liens from these studies and case studies. I compiled the 2000 and 2002 data by contacting state Medicaid offices and talking to local practitioners in states.

32  

References

Brown, Jeffrey R. and Amy Finkelstein, “Why is the Market for Long-Term Care Insurance so Small?” Journal of Public Economics 91(2007): 1967-1991.

Brown, Jeffrey R. and Amy Finkelstein, “The Interaction of Public and Private Insurance:

Medicaid and the Long-Term Insurance Market.” American Economic Review 98(2008): 1083-1102.

Brown, Jeffrey R., Norma B. Coe, and Amy Finkelstein, “Medicaid Crowd-Out of

Private Long Term Care Insurance Demand: Evidence from the Health and Retirement Survey.” Tax Policy and the Economy 21(2007): 1-34.

Cameron, A. Colin, Jonah Gelbach, and Douglas Miller. “Robust Inference with Multi

Way Clustering” NBER Working Paper No. T0327, 2006. Centers for Medicare and Medicaid Services

http://www.cms.hhs.gov/NationalHealthExpendData/ Coe, Norma B., “Financing Nursing Home Care: New Evidence from Spend Down

Behavior” Mimeo., Tilburg University, 2007. Congressional Research Service, Background Data and Analysis in Medicaid Source

Book, Washington, D.C.: U.S. Government Printing Office, 1993. Costa, Dora L., “A House of Her Own: Old Age Assistance and Living Arrangements of

Older Nonmarried Women,” Journal of Public Economics, 72 (1999): 39-60. Davidoff, Thomas, “Maintenance and the Home Equity of the Elderly,” Mimeo., UC

Berkeley, 2004. Dick, Andrew, Alan M. Garber, and Thomas A. MaCurdy, in David Wise (ed.) Studies in

the Economics of Aging, 1994. Dietz, Robert D. and Donald R. Haurin, “The Social and Private Micro-Level

Consequences of Homeownership,” Journal of Urban Economics 54 (2003): 401-450. Engelhardt, Gary V., “Social Security and Elderly Homeownership,” Journal of Urban

Economics 63 (2008): 280-305. Engelhardt, Gary V., and Nadia Greenhalgh-Stanley, “Public Long-Term Care Insurance

and the Housing and Living Arrangements of the Elderly: Evidence from Medicare Home Health Benefits,” Mimeo., Syracuse University 2008.

Engelhardt, Gary V., Jonathon Gruber, and Cynthia D. Perry, “Social Security and

Elderly Living Arrangements,” Journal of Human Resources 40:3 (2005): 354-372.

33  

Gruber, Jonathon. 2003. “Medicaid,” in Robert Moffitt, ed., Means Tested Transfer

Programs in the U.S. Chicago: University of Chicago Press, pp. 15-77. Hurst, Erik, and James P. Ziliak. 2006. “Do Welfare Asset Limits Affect Household

Saving? Evidence from Welfare Reform.” Journal of Human Resources 41 (1): 46-71. Karp, Naomi, Charles Sabatino, and Erica Wood. 2005. “Medicaid Estate Recovery: A

2004 Survey of State Programs and Practices.” The Public Policy Institute, AARP. June. Krueger, Alan B. and Bruce D. Meyer, “Labor Supply Effects of Social Insurance,” in

Alan Auerbach and Martin Feldstein (eds.) Handbook of Public Economics, Volume 4 (Amsterdam: North Holland), 2002.

Long Term Care Link – State Medicaid Estate Recovery Programs

http://www.longtermcarelink.net/reference/ref_medicaid_recovery.html  McGarry, Kathleen, and Robert F. Schoeni, “Social Security, Economic Growth, and the

Rise in Elderly Widows’ Independence in the Twentieth Century,” Demography, 37 (2000), 221-236.

Michael, Robert T., Victor R. Fuchs, Sharon R. Scott. 1980. “Changes in the Propensity

to Live Alone,” Demography, 17(1980): 39-56. Norton, Edward C., “Long-Term Care,” in A.J. Cuyler and J.P. Newhouse (eds.)

Handbook of Health Economics, Volume 1 (Elsevier: Amsterdam), 2000, pp. 955-994. Office of Federal Housing Enterprise Oversight http://www.ofheo.gov/HPI.aspx. Poterba, James, “Estate and Gift Taxes and Incentives for Inter Vivos Giving in the

United States,” Journal of Public Economics 79(2001): 237-264. Sabatino, Charles P. and Erica Wood. “Medicaid Estate Recovery: A Survey of State

Programs and Practices”. Public Policy Institute#9615 September 1996. Schwartz, Roger A. and Charles P. Sabatino. “Medicaid Estate Recovery Under OBRA

’93: Picking the Bones of the Poor?” Prepared for Commission on Legal Problems of the Elderly November 1994.

Taylor, Donald, Frank Sloan, and Edward Norton. 1999. “Formation of Trusts and

Spend Down to Medicaid.” Journal of Gerontology:Social Sciences 54(B). Venti, Steven F., and David A. Wise, “Aging and Housing Equity,” NBER Working

Paper No.7882, 2000. Wood, Erica and Ellen Klem. 2007. “Protections in Medicaid Estate Recovery:

Findings, Promising Practices, and Model Notices.” ABA Commission on Law

34  

and Aging.

35  

Table 1. AHEAD Cohort Sample Statistics

Note: The portfolio share is the percent of total wealth comprised of home equity. All home equity and total wealth measures are reported in 2005 adjusted real dollars. Data from the AHEAD cohort of the HRS are used to obtain these sample statistics. Nursing Home (%) describes the percent of the sample who reported they spent time in a nursing home in the past two years. Medicaid (%) is the proportion of the sample that responded they received Medicaid.  

  

(1) (2) (3) (4) (5) (6) (7)

Sample Restrictions

All

Married

Unmar

TEFRA

Non-TEFRA

ERP

Non-ERP

Home Ownership Rate(%)

89.4 95.1 82.9 87.0 90.2 88.9 91.8

Live in TEFRA states(%)

27.3 26.6 28.1 100 0.00 36.5 0.00

Live in ERP states(%)

74.8 74.9 74.7 100 65.3 100 0.00

Housing Portfolio Share(%)

54.9 53.4 56.5 54.7 54.9 53.6 58.6

Trust Alive(%) 13.5 15.0 11.7 18.4 11.6 15.0 8.8 Home Ownership at Death(%)

59.9 62.3

56.5 57.6 59.7 58.8 60.4

Trust at Death(%) 15.7 15.9 15.9 24.3 12.6 17.9 9.2 Mean Home Equity 119,124 138,082 97,278 160,447 103,623 127,957 92,935 Median Home Equity 84,690 97,182 5,332 108,837 77,746 88,430 74,826 Nursing Home (%) 6.9 3.5 10.8 7.9 6.5 7.4 5.3 Medicaid (%) 6.7 5.2 10.3 5.8 8.0 6.0 11.6 Sample Size 24584 13108 11476 6886 17698 18379 6205

36  

Table 2. Linear Probability Estimated Impact of TEFRA Liens and Estate Recovery Programs on Home Ownership Decisions at the Time of Death, (Standard Errors in Parentheses)

(1) (2) (3) (4) (5) (6)

Own Home at Death Without Controls

Without Controls

With Controls

Housing Price Index

State Medicaid Expend

Linear State by

Time Trends

ERP*Unmarried -0.155 (0.053)

[-25.8%]

-0.131 (0.058)

[-21.8%]

-0.119 (0.060)

[-19.8%]

-0.118 (0.060)

[-19.7%]

-0.119 (0.060)

[-19.8%]

-0.126 (0.060)

[-21.0%] TEFRA*Unmarried

-0.065 (0.055)

[-10.8%]

-0.076 (0.059)

[-12.7%]

-0.077 (0.060)

[-12.8%]

-0.080 (0.057)

[-13.3%]

-0.079 (0.053)

[-13.2%] ERP

0.091

(0.054) [15.2%]

0.085

(0.054) [14.2%]

0.111

(0.053) [18.5%]

0.111

(0.054) [18.5%]

0.122

(0.051) [20.3%]

0.188

(0.061) [31.3%]

TEFRA

-0.017 (0.045) [-2.8%]

-0.013 (0.049) [2.2%]

-0.014 (0.049) [2.3%]

-0.026 (0.056) [-4.3%]

-0.024 (0.046) [-4.0%]

Minority

0.093

(0.031) [15.5%]

0.093

(0.031) [15.5%]

0.093

(0.031) [15.5%]

0.090

(0.030) [15.0%]

Kids

-0.002 (0.004) [-0.3%]

-0.002 (0.004) [-0.3%]

-0.002 (0.004) [-0.3%]

-0.0007 (0.004) [-0.1%]

College

0.042

(0.028) [7.0%]

0.042 (0.028) [7.0%]

0.042 (0.028) [7.0%]

0.037 (0.031) [6.2%]

Mean Home Ownership(%) Number of Observations

59.9

3118

59.9

3118

59.9

3118

59.9

3118

59.9

3118

59.9

3118 Joint Test of Interactions P-value

10.2 0.006

9.51 0.009

9.65 0.008

10.0 0.007

13.26 0.001

Note: This table shows the estimated impact of living in a state with TEFRA liens and ERPS on home ownership rates at death, using OLS estimation. The excluded group consists of white males with less than a college education. A dummy variable for female, unmarried, and age are included in the regression analysis for columns (3) – (6) but not shown in the table. A full set of fixed state and year effects are also included in all columns. Standard errors clustered by state and person identifier (using the Cameron, Gelbach, and Miller multi-dimension clustering technique) are presented in parentheses and marginal effects are shown in square brackets, which are the percentage point changes divided by the sample homeownership rate to get the percent changes. Column (4) adds a housing price index to the estimation. Column (5) includes a measure of state Medicaid expenditure, presented in real 2005 dollars. Column (6) augments the estimation by including a state-by-time linear trend into the regression. The chi-squared value from the joint hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability.

37  

Table 3. Linear Probability Estimated Impact of TEFRA Liens and Estate Recovery Programs on Trust Participation at the Time of Death, (Standard Errors in Parentheses) (1) (2) (3) (4) (5) (6)

Trust at Death Without Controls

Without Controls

With Controls

Housing Price Index

State Medicaid Expend

Linear State by

Time Trends

ERP*Unmarried 0.041 (0.019) [26.1%]

0.010 (0.016) [6.4%]

0.015 (0.018) [9.6%]

0.016 (0.019) [10.2%]

0.015 (0.018) [9.5%]

0.013 (0.022) [8.3%]

TEFRA*Unmarried

0.085

(0.029) [54.1%]

0.077

(0.028) [49.0%]

0.073

(0.028) [46.5%]

0.073

(0.025) [46.5%]

0.058

(0.022) [36.9%]

ERP

-0.085 (0.044)

[-54.1%]

-0.069 (0.037)

[-43.9%]

-0.068 (0.036)

[-43.3%]

-0.070 (0.036)

[-44.6%]

-0.056 (0.033)

[-35.7 %]

-0.079 (0.053)

[-50.3%] TEFRA

-0.042 (0.042)

[-26.8%]

-0.045 (0.044)

[-28.7%]

-0.048 (0.043)

[-30.6%]

-0.061 (0.054)

[-38.9%]

-0.058 (0.040)

[-36.9%] Minority

-0.070 (0.023)

[-44.6%]

-0.070 (0.023)

[-44.6%]

-0.070 (0.022)

[-44.6%]

-0.070 (0.023)

[-44.6%] Kids

-0.006 (0.004) [-3.8%]

-0.006 (0.004) [-3.8%]

-0.006 (0.004) [-3.8%]

-0.006 (0.004) [-3.8%]

College

0.111

(0.025) [70.7%]

0.111 (0.025) [70.7%]

0.111 (0.025) [70.7%]

0.108 (0.027) [68.8%]

Mean Trust Participation (%) Number of Observations

15.7

3118

15.7

3118

15.7

3118

15.7

3118

15.7

3118

15.7

3118 Joint Test of Interactions P-value

12.05 0.002

12.15 0.002

12.12 0.002

14.42 0.0007

14.36 0.0008

Note: This table shows the estimated impact of living in a state with TEFRA liens and ERPs on trust participation at death, using OLS estimation. The excluded group consists of white males with less than a college education. A dummy variable for female, unmarried, and age are included in the regression analysis for columns (3) – (6) but not shown in the table. A full set of fixed state and year effects are also included in all columns. Standard errors clustered by state and person identifier (using the Cameron, Gelbach, and Miller multi-dimension clustering technique) are presented in parentheses and marginal effects are shown in square brackets, which are the percentage point changes divided by the sample mean to get the percent changes. Column (4) adds a housing price index to the estimation. Column (5) includes a measure of state Medicaid expenditure, presented in real 2005 dollars. Column (6) augments the estimation by including a state-by-time linear trend into the regression.   The chi-squared value from the joint hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability.

38  

Table 4 Time Trends – Unbalanced Panel Estimated Impact of TEFRA Liens and Estate Recovery Programs (ERPs) on Home Ownership Decisions at death, interview prior to death, two interviews prior to death, and while alive, (Standard Errors in Parentheses)

(1) (2) (3) (4)

Own Home At Death Interview Prior to Death

Two Interviews Prior to Death

While Alive

TEFRA*Unmarried

-0.076 (0.059)

[-12.7%]

-0.011 (0.047) [-1.5%]

-0.015 (0.034) [-1.8%]

-0.024 (0.014) [-2.7%]

ERP*Unmarried

-0.119 (0.060)

[-19.8%]

-0.029 (0.026) [-3.9%]

-0.026 (0.030) [-3.1%]

-0.016 (0.018) [-1.8%]

Minority

0.093

(0.031) [15.5%]

0.081

(0.029) [10.8%]

0.040

(0.021) [4.8%]

0.033

(0.008) [3.7%]

Kids

-0.002 (0.004) [-0.3%]

0.0002 (0.004)

[-0.002%]

-0.002 (0.004) [-0.2%]

-0.003 (0.002) [-0.2%]

College

0.042

(0.028) [7.0%]

0.023

(0.023) [3.1%]

0.009

(0.015) [1.1%]

0.004

(0.007) [0.6%]

Mean Home Ownership(%) Number of Observations

59.9

3118

75.0

2636

84.1

2589

89.4

24457 Joint Test of Significance 9.51

2.04 0.79 6.90

P-Value 0.009 0.360 0.674 0.032 Note: This table shows the estimated impact of living in a state with TEFRA liens on home ownership rates at death, 0-2 years before death, 2-4 years before death, and while alive to show a time trend of the impact of these laws. The excluded group consists of white males with less than a college education. A dummy variable for female, unmarried, and the relevant demographics are also included in the analysis but are not shown in the table. A full set of state and year fixed effects are included in the estimation. Standard errors clustered by state and person identifier are shown in parentheses and marginal effects are shown in square brackets. The chi-squared value from the hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability.

39  

Table 5 Time Trends – Balanced Panel Estimated Impact of TEFRA Liens and Estate Recovery Programs (ERPs) on Home Ownership Decisions at death, interview prior to death, two interviews prior to death, (Standard Errors in Parentheses)

(1) (2) (3)

Own Home At Death Interview Prior to Death

Two Interviews Prior to Death

TEFRA*Unmarried

-0.026 (0.052) [-4.2%]

-0.004 (0.047) [-0.5%]

-0.015 (0.034) [-1.8%]

ERP*Unmarried

-0.027 (0.060) [-4.3%]

-0.032 (0.028) [-4.3%]

-0.026 (0.030) [-3.1%]

Minority

0.123

(0.032) [19.8%]

0.086

(0.027) [11.5%]

0.040

(0.021) [4.8%]

Kids

-0.002 (0.004) [-0.3%]

-0.00003 (0.004)

[-0.004%]

-0.002 (0.004) [-0.2%]

College

0.044

(0.029) [7.1%]

0.023

(0.023) [3.1%]

0.009

(0.015) [1.1%]

Mean Home Ownership(%) Number of Observations

62.2

2589

75.1

2589

84.1

2589 Joint Test of Significance 0.76 1.66 0.79 P-Value 0.684 0.437 0.674

Note: This table shows the estimated impact of living in a state with TEFRA liens on home ownership rates at death, 0-2 years before death, and 2-4 years before death to show a time trend of the impact of these laws. The excluded group consists of white males with less than a college education. A dummy variable for female, TEFRA, unmarried, ERP, and the relevant demographic variables are also included in the analysis but are not shown in the table. A full set of state and year fixed effects are included in the estimation. Standard errors clustered by state and household identifier are shown in parentheses and marginal effects are shown in square brackets. The chi-squared value from the hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability. This sample was limited to those alive 2-4 years prior to death in the HRS data set, so as to follow the same sample over time.

40  

Table 6. OLS Estimated Impact of TEFRA liens and Estate Recovery Programs (ERPs) on Home Equity Decisions while alive, (Standard Errors in Parentheses)

(1) (2) (3) (4) (5) (6)

Home Equity Without Controls

Without Controls

With Controls

Housing Price Index

State Medicaid Expend

Linear State by Time Trends

ERP*Unmarried -3674.1 (8675.0) [-3.1%]

-359.82 (8868.3) [-0.3%]

-2900.1 (8926.5) [-2.4%]

-3170.0 (8467.5) [-2.7%]

-3360.3 (8546.3) [-0.3%]

-2860.5 (8962.9) [-2.4%]

TEFRA*Unmarried

-18559 (13173) [-15.6%]

-24813 (13532) [-20.8%]

-22320 (12836) [-18.7%]

-22000 (12903) [-18.5%]

-20407 (13489) [-17.1%]

ERP

140.88

(9563.8) [0.1%]

-1208.2 (81228) [-1.0%]

279.31

(7951.8) [0.2%]

-74.043 (8533.5) [-0.06%]

1177.5

(7855.9) [1.0%]

9192.1 (12827) [7.7%]

TEFRA

6966.3

(9246.2) [5.8%]

8113.0

(9200.8) [6.8%]

6301.4

(9461.5) [5.3%]

6615.1

(9296.4) [5.6%]

3053.7

(9434.0) [2.6%]

Minority

-37064

(7111.6) [-31.1%]

-37094

(7093.4) [-31.1%]

-37068

(7111.6) [-31.1%]

-37940

(6903.2) [-31.8%]

Kids

747.17

(1279.0) [0.6%]

740.35

(1278.6) [0.6%]

745.47

(1278.5) [0.6%]

977.23

(1413.8) [0.8%]

College

51212

(8992.9) [43.0%]

51222

(8997.0) [43.0%]

51222

(8992.2) [43.0%]

51471

(9072.6) [43.2%]

Mean Home Equity Number of Observations

119,124

24457

119,124

24457

119,124

24457

119,124

24457

119124

24457

119124

24457 Joint Test of Significance 3.92 4.41 4.57 4.49 3.24 P-Value 0.141 0.110 0.102 0.106 0.198

Note: This table shows the estimated impact of living in a state with TEFRA liens on home equity decisions while alive. The excluded group consists of white males with less than a college education. A dummy variable for female, TEFRA, unmarried, ERP, and the relevant demographics are also included but are not shown in the table. A full set of state and year fixed effects are included in the estimation. Standard errors clustered by state and person identifier are shown in parentheses using the Cameron, Gelbach, and Miller estimation technique and marginal effects are shown in square brackets. The chi-squared value from the hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability. Column (4) adds in housing price index, column (5) adds in state Medicaid expenditure, and column (6) adds in linear state-by-time trends.

41  

Table 7. Linear Probability Estimated Impact of TEFRA liens and Estate Recovery Programs (ERPs) on Portfolio Decisions while alive, (Standard Errors in Parentheses)

(1) (2) (3) (4) (5) (6)

Portfolio Share Without Controls

Without Controls

With Controls

Housing Price Index

State Medicaid Expend

Linear State by Time Trends

ERP*Unmarried -0.039 (0.031) [-7.1%]

-0.008 (0.028) [-1.5%]

-0.002 (0.029) [-0.4%]

-0.003 (0.029) [-0.6%]

-0.002 (0.029) [-0.4%]

-0.003 (0.028) [-0.6%]

TEFRA*Unmarried

-0.083 (0.039)

[-15.1%]

-0.081 (0.035)

[-14.8%]

-0.080 (0.034)

[-14.6%]

-0.082 (0.036)

[-14.9%]

-0.081 (0.041)

[-14.8%] ERP

0.003

(0.030) [-0.5%]

-0.010 (0.031) [-1.8%]

-0.012 (0.031) [-2.2%]

-0.012 (0.031) [-2.2%]

-0.010 (0.030) [-1.8%]

-0.031 (0.059) [-5.6%]

TEFRA

0.038

(0.032) [6.9%]

0.039

(0.030) [7.1%]

0.042

(0.032) [7.7%]

0.035

(0.024) [6.4%]

0.066

(0.046) [12.0%]

Minority

0.180

(0.021) [32.8%]

0.180

(0.021) [32.8%]

0.180

(0.021) [32.8%]

0.180

(0.021) [32.8%]

Kids

0.016

(0.005) [2.9%]

0.016

(0.005) [2.9%]

0.016

(0.005) [2.9%]

0.016

(0.005) [2.9%]

College

-0.112 (0.010)

[-20.4%]

-0.112 (0.010)

[-20.4%]

-0.112 (0.010)

[-20.4%]

-0.112 (0.010)

[-20.4%] Mean Portfolio Share(%) Number of Observations

54.9

24457

54.9

24457

54.9

24457

54.9

24457

54.9

24457

54.9

24457 Joint Test of Significance 5.26 5.69 5.77 5.30 4.09 P-Value 0.072 0.058 0.056 0.071 0.129

Note: This table shows the estimated impact of living in a state with TEFRA liens on housing as a proportion of total wealth while alive. The excluded group consists of white males with less than a college education. A dummy variable for female, TEFRA, unmarried, ERP, and the relevant demographics are also included but are not shown in the table. A full set of state and year fixed effects are included in the estimation. Standard errors clustered by state and person identifier are shown in parentheses using the Cameron, Gelbach, and Miller estimation technique and marginal effects are shown in square brackets. The chi-squared value from the hypothesis that the coefficient on unmarried*TEFRA and unmarried*ERP are both zero is reported at the bottom of the table as well as the associated probability. Column (4) adds in housing price index, column (5) adds in state Medicaid expenditure, and column (6) adds in linear state-by-time trends.