The Household Bankruptcy Decision
By SCOTT FAY, ERIK HURST, AND MICHE LLE J. WHITE*
Personal bankruptcy lings have risen from
0.3 percent of households per year in 1984 to
around 1.35 percent in 1998 and 1999, trans-
forming bankruptcy from a rare occurrence to a
routine event. Lenders lost about $39 billion in
1998 due to personal bankruptcy lings.
1
But
economists have little understanding of why
households le for bankruptcy or why lings
have increased so rapidly. Until very recently,
studying the household bankruptcy decision
was very dif cult, because no household-level
data set existed that i ncluded information on
bankruptcy lings. In this paper, we use new
data from the Panel Study of Income Dynamics,
which includes information on bankruptcy l-
ings, to estimate a model of households’ bank-
ruptcy decisions.
We nd support for the strategic model o f
bankruptcy, which predicts that households are
more likely to le when their nancial bene t
from ling is higher. Our model predicts that an
increase of $1,000 in h ouseholds’ nancial ben-
e t from bankruptcy would result in a 7-p ercent
increase in the number of bankruptcy lings.
Our model also predicts that if the 1997 Na-
tional Bankruptcy Review Commission’s pro-
posed changes i n bankruptcy exemption levels
were implemented, there would be a 1 6-percent
increase in the number of bankruptcy lings
each year. But if the $100,000 cap on home-
stead exemptions recently passed by the U.S.
Senate were adopted, our model predicts that
there would be only a negligible effect on the
number of lings. We nd little support for the
nonstrategic model of bankruptcy which pre-
dicts that households le when adverse events
occur which reduce their ability to repay. Fi-
nally, contro lling for state and time xed ef-
fects, our mod el shows that households are
more likely to le for bankruptcy if they live in
districts with higher aggregate ling rates.
I. U. S. Personal Bankruptcy Law
The United States has two different personal
bankruptcy procedures—Chapter 7 and Chapter
13—and debtors have the right to choose be-
tween them.
Chapter 7.
—Under Chapter 7, unsecured
debts such as credi t ca rd debt, installment loans,
medical bills, and damage claims are dis-
charged. Debtors are not obliged to use any of
their future earnings to rep ay their debt, but they
are obliged to turn over all of their as sets above
a xed exemption level to the bankruptcy
trustee. T he trustee liquidates the nonexempt
assets and uses the proceeds to repay creditors.
Although bankruptcy is a matter of federal law
and the rules are uniform across the United
States, Congress gave the sta tes the ri ght to
adopt their own bankruptcy ex emptions. Most
states have separate exemptions for equity in the
debtor’s principle residence (the “homestead
exemption”) and for several types of personal
property. In general, states’ nonhomestead ex-
emptions are low, but their homestead exemp-
tions vary widely, from a few thousand dollars
to unlimited in nine states.
2
If debtors’ assets
are less than the exemption levels in their states,
then they are not obliged to repay anything to
creditors.
* Fay: Warrington College of Business Administration,
University o f Florida; Hurst: Graduate School of Business,
University of Chicago; White: Department of Economics,
University o f California, San Diego, and NBER. We are
grateful to Orley Ashenfelter, John Bound, Charlie Brown,
Paul Courant, Austan Goolsbee, Jim Hines, Joe Lupton,
Ariel Pakes, Gary Solon, Nicholas Souleles, Frank Stafford,
Elizabeth Warren, and an anonymous referee for h elpful
comments, and to Citibank and the National Science Foun-
dation, under Grant No. SBR 96–17712, for nancial sup-
port. Earlier versions of this paper were presented at
Harvard University, the Wharton School, the University of
Chicago, UCLA, and the American Law and Economics
Association Conference.
1
This gure is based on unsecured debt per bankruptcy
ling of $28,000 (John M. Barron and Michael Staten,
1998).
2
The average value of all nonhomestead exemptions in
1995 was $5,000. The average homestead exemption in
1995 for states that do not have u nlimi ted homestead ex-
emptions was $25,000. Most states also exempt clothing,
furniture, and h ouseh old goods.
706
Households’ nancial bene t from ling for
bankruptcy under Chapter 7 is therefore the
value of debt discharged and their nancial cost
is the value o f nonexempt assets, if any, that
they must give up. Households’ net nancial
bene t from ling for bankruptcy is the differ-
ence. Households that le for bankruptcy must
also pay bankruptcy court ling fees and law-
yers’ fees. They also face possible nonpecuni-
ary costs, including the cost of acquiring
information about the bankruptcy pro cess,
higher future borrowing costs, and the cost of
bankruptcy stigma.
Chapter 13.
—Chapter 13 bankruptcy is in-
tended for debtors who earn regular incomes.
Under it, debtors do not give up any asse ts in
bankruptcy, but they must propose a plan to
repay a portion of their debts from future in-
come, usually over three to ve years. The plan
goes into effect as long as the bankruptcy judge
accepts it, i.e., creditors do not have the right to
block repayment p lans.
Because debtors have the right to choose
between Chapters 7 and 13, they have a nan-
cial incentive to choose Chapter 7 whenever
their assets are less than their state’s exemption,
since doing so allows them to avoid repaying
their debts completely.
3
Even when households
le under Chapter 13, they are obliged to use
future earnings to repay debt only to the extent
that they would be obliged to use nonexempt
assets to repay debt under Chapter 7. For exam-
ple, debtors who h ave $5,000 in nonexempt
assets are obliged to repay o nly the equivalent
of $5,000 from future earnings in a repayment
plan under Chapter 13. Debtors who have no
nonexempt as sets sometimes le under Chapter
13, but propose to repay only token amounts.
Bankruptcy judges vary in their willingness to
accept these plans.
4
II. Literature Review
Attempts to study the bankruptcy ling deci-
sion have been hampered by the lack of
household-level data on bankruptcy lings. In
an early study, White (1987) regressed the ag-
gregate bankruptcy ling rate by county on the
bankruptcy exemption level for the relevant
state and o ther variables. She fo und that the
bankruptcy ling rate was positively and sig-
ni can tly related to the exemption level.
5
Ian Domowitz and Robert L. Sartain (1999)
got around the lack of household-level data
on bankruptcy lings by combining two data
sources: a sample of households that led for
bankruptcy under Chapter 7 in the early 1980’s
and a representative sample of U.S. households
which includes detailed nancial information
(the 1983 Survey of Consumer Finances). They
found that h ouseholds with more credit card
debt were more likely to le for bankruptcy.
David Gross and Nicholas Souleles (2002) used
a dat a set of i ndividual credit card accounts to
explain account holders’ b ankruptcy decisions.
Their main explanatory variable is lenders’ rat-
ing of individual account holders’ riskiness and
their main nding is that, after controll ing for
the increase in the average borrower’s riskiness,
the probability of default rose signi cantly be-
tween 1995 and 1997. They interpret this result
as evidence that the level o f bankruptcy stigma
has fallen. Neither the Do mowitz and Sartain
nor the Gross and Souleles papers tested
whether households’ decisions to le for bank-
ruptcy are related to their nancial bene t from
ling, which is a central goal of this study.
6
3
Debtors may shift assets from nonexempt to exempt
categories before ling or use other strategies to reduce their
nonexempt assets before ling. Debtors may also default on
their debt but not le for bankruptcy, since creditors do not
always attempt to collect. See White (1998a) for discussion.
4
About 70 percent of bankruptcy lings occur under
Chapter 7. Congress has attempted to make Chapter 13
more attractive to debtors by allowing some types of
debts—including some student loans and debts incurred by
fraud—to be discharged under Chapter 13, but not under
Chapter 7. Debtors are also allowed to le under Chapter 13
as often as every six months, while they cannot le under
Chapter 7 more often than once every six years. Chapter 13
is also attractive to debtors who own homes and are in
arrears on mortgage payments, because it delays foreclo-
sure. In 1984, Congress adopted a provision intended to
prevent high-income debtors from ling under Chapter 7,
but later court decisions and lack of enforcement made it
ineffective. See Karen Gross (1986), Wayne R. Wells et al.
(1991), and White (1998b) for discussions of this provision
and the relationship between Chapters 7 and 13.
5
Frank H. Buckley and Margaret F. Brinig (1998) did a
similar st udy using state rather than county bankruptcy
ling rates, for the years 1980 to 1991. They found a
negative relationsh ip between state aggregate ling rates
and the exemption level.
6
For theoret ical models of the bankruptcy decision, see
Samuel A. Rea, Jr. (1984) and Ronald A. Dye (1986).
Buckley (1994) discusses explanations for the pro-debtor
tilt of U.S. bankruptcy policy. Reint Gropp et al. (1997)
707VOL. 92 NO. 3 FAY ET AL.: THE HOUSEHOLD BANKRUPTCY DECISION
There is also a sociologically oriented litera-
ture on the bankruptcy ling decision. Teresa A.
Sullivan et al. (1989) examined the characteris-
tics of a sample of households that led for
bankruptcy during t he early 1980’s. Based on
descriptive evidence, they argued that house-
holds le for bankruptcy whe n unexpected ad-
verse events occur which reduce their ability to
repay their debts. Sullivan et al. also argue that
households do not take nancial be ne t into
account in making their bankruptcy decisions.
We test the adverse events hypothesis in our
empirical work.
7
Finally, evidence from several sources sug-
gests that the administration and p ractice of
bankruptcy law vary across bankruptcy dis-
tricts, which may cause incentives to le for
bankruptcy to vary across districts. Jean
Braucher (1993) interviewed bankruptcy law-
yers in four bankruptcy districts and found that
they often discourage debtors who have less
than a minimum amount of dischargeable debt
from ling for bankruptcy, but the minimum
amount varies across districts. Braucher also
notes that bankruptcy trustees in each district set
standard legal fees for Chapter 7 and Chapter 13
bankruptcy lings. Because these fees vary
widely across districts, lawyers’ incentives to
specialize in bankruptcy cases also vary across
districts. Both Braucher and Sullivan et al.
(1989) h ave noted that there are large variations
across bankruptcy districts in the proportion of
lings that o ccur u nder Chapter 13, which they
attribute to judges or lawyers in particular dis-
tricts encou raging debtors to le under Chapter
13. But pressure to le under Chapter 13 could
make ling for bankruptcy either more or less
attractive overall, depending on whether bank-
ruptcy judges in the district are willing to accept
token repayment plans under Chapter 13. In our
empirical work, we test whether the individual
households’ decisions t o le for bankruptcy are
in ue nced by the number of bankruptcy lin gs
in their districts.
III. Data and Speci cation
In 1996, th e Panel Study of Income Dynam-
ics (PSID) asked respondents whether they had
ever led for bankruptcy and, if so, in what
year(s). Our data set is a combined cross-
section, time-series sample of PSID households
in the years 1984–1995. We run probit regres-
sions explaining whether household
i
led for
bankruptcy in year
t
.
8
The independent variables test three hypoth-
eses: whether households are more likely to le
for bankru ptcy as their net nancial bene t from
ling increases, whether (controlling for nan-
cial bene t) they are more likely to le for
bankruptcy when adverse events occur, and
whether households’ bankruptcy decisions are
in ue nced by average bankruptcy ling rates in
the localities where they live.
A.
Financial Bene t
Consider rst the hypothesis that households
are more likely to le for ban kruptcy as their net
nancial bene t from ling increases. As dis-
cussed above, household
i
’s net nancial bene t
from ling for Chapter 7 bankruptcy in year
t
is:
(1)
FinBen
it
5
max
@D
it
2
max
@W
it
2 E
it
, 0
#
, 0
#
where
D
it
is the value of household
i
’s unse-
cured debt that would be discharged in bank-
ruptcy in year
t
,
W
it
is household
i
’s wealth in
investigate the effect of variations in bankruptcy exemp-
tions on supply and demand for consumer credit.
7
A Washington Post (February 18, 2000) editorial argu-
ing against changing current bankruptcy law suggests that
this i s a commonly held view: “Most bankruptcies are
triggered by misfortune, not irresponsibility: by illness, a
job loss, a broken marriage. America should remain the
home of second chances.”
8
In order for particular households to be included in our
sample, they must have answered all of the PSID question-
naires for the years 1992–1995. Households that are in the
sample for 1992–1995 are also included for any of th e
additional years 1984–1991 for which data are available.
We used a balanced panel for the years 1992–1995 because
the PSID data sets for 1993–1996 are only available in
“early release” form and no household weights are included.
We therefore used 1992 household weights for all of the
1993–1995 observations. The “early release” data sets also
omit households’ state of residence. As a result, we are
forced t o assume that households observed in 1993–1995
still live in the same state where they lived in 1992. We used
the con den tial PSID geocodes to assign households to their
counties of residence in each year of the sample (up to
1992). This allows us to assign households to bankruptcy
districts and also to use county-level data for the unemploy-
ment rate. Because we use the PSID weights, our sample is
representative of the general population.
708 THE AMERICAN ECONOMIC REVIEW JUNE 2002
year
t
net of secured debts such as mortgages
and car lo ans, and
E
it
is the bankruptcy exemp-
tion in household
i
’s state of residence in year
t
.
When household
i
les for bankruptcy, debts of
D
it
are discharged, but the household must give
up assets of value
W
it
2 E
it
if its wealth
W
it
exceeds the exemption level
E
it
.
FinBen
it
must
be nonnegative, since households would not le
for bankruptcy if their nonexempt assets ex-
ceeded the amount of debt discharged. Al-
though equation (1) gives th e nancial bene t
of l ing under Chapter 7, it als o applies to l-
ing under Chapter 13, because—as discussed
above— households have a choice between the
two procedures and their nancial bene t from
ling under Chapter 13 is closely related to their
nancial bene t from ling under Chapter 7.
To calculate nancial bene t, we obtained
exemption levels by stat e from 1984–1995 for
equity in owner-occupied homes, equity in ve-
hicles, and personal property applicable to -
nancial assets, plus the wild card exemption
(which can be applied to any asset). The bank-
ruptcy exemption variable
E
it
is assumed to
equal the sum of these exemptions if the house-
hold owns its own home or the sum of the
vehicle, personal property, and wild card ex-
emptions if the household rents. Since most
states allow married couples who le for bank-
ruptcy to t ake hig her exemptions, we also adjust
the exemption levels by the appropriate amount
if the household contains a married couple. If
the state’s homestead exemption is unlimited
and th e household owns its own home, we as-
sume that the value of the homestead exemption
equals the value of the household’s home.
9
Six-
teen states also allow their residents to choose
between the state’s exemption and a uniform
federal bankruptcy exemption. For residents of
these states, we use the larger of the state or the
federal exemption.
10
The other variables needed t o calculate net
nancial bene t,
D
it
and
W
it
, are taken from the
PSID. The PSID asks questions concerning the
amount of unsecured debt and the value of
nonhousing wealth only as part of the wealth
supplements, which were conducted in 1984,
1989, and 1994, but it asks the value of housing
equity every year. We use 1984, 1989, and 1 994
data o n unsecured debt to construct
D
it
for each
of the years 1984–1988, 1989–1993, and
1994–1995, respectively. Household
i
’s wealth
in year
t
,
W
it
, equals the value of housing
equity in year
t
plus the value of nonhousing
assets from the most recent wealth survey prior
to year
t
. The fact that data on unsecured d ebt
and n onhousing assets are only available in
ve-year increments means that o ur measure of
nancial bene t is subject to measurement
error.
11
We include b oth nancial bene t and nan -
cial be ne t squared as regressors in our model
of the bankruptcy ling decision in order to test
for potential nonlinearities in the effect of -
nancial bene t on th e bankruptcy decision.
B.
Adverse Events
The nonstrategic view of bankruptcy is that
households do not plan in advance for bank-
ruptcy and do not respond to nancial gain in
deciding whether to le. Instead, they le in
response to unanticipated adverse events which
reduce their ability to repay their debts. We
would like to test the nonstrategic model of
bankruptcy against the strategic model just dis-
cussed. A strict interpretation of the nonstrate-
gic model implies that income should be
negatively and signi cantly related to the prob-
ability of ling for bankruptcy, because income
measures ability to repay debt. B ut nancial
bene t should not be signi cantly related to the
probability of ling, because households’ nan -
cial bene t from ling depends only on their
wealth and not on their incomes. Conversely, a
strict interpretation of the strategic model implies
9
This assumes that households take advantage of the
various bankruptcy exemptions by converting assets from
nonexempt to exempt categories where possible.
10
In the 1978 Bankruptcy Code, Congress adopted a
uniform federal bankruptcy exemption, bu t permitted states
to opt out of the federal exemption by ado pting their own
exemptions. All states had done so by 1983, but about
one-third of the states allow their residents to choose be-
tween the state and the federal exemptions. Since the early
1980’s, the pattern has been that states change their exemp-
tion levels only rarely—mainly to correct nominal exemp-
tion levels fo r in ation. Because of this, we treat th e
exemption levels as exogenous.
11
See our working pap er, Fay et al. (1998), for discus-
sion o f how measurement error might bias our ndings of
the marginal effect of changes in nancial bene t on house-
holds’ probability of ling for bankruptcy and a test for the
effect of measurement error. We nd that measurement
error does not signi cantly affect our results.
709VOL. 92 NO. 3 FAY ET AL.: THE HOUSEHOLD BANKRUPTCY DECISION
that the nancial bene t variable should be pos-
itively and signi cantly related to the probabil-
ity of ling for b ankruptcy, but income should
not, because income is unrelated to the nancial
gain from bankruptcy. Thus a regression of income
and nancial bene t on whether households le
for bankruptcy should allow us to distinguish
between the theories.
However, mismeasurement of wealth is
likely to pre vent us from cleanly distinguishing
between the two theories. As discussed above,
our measure of nancial bene t relies on wealth
data wh ich is only collected at ve-year incre-
ments. S ince current income acts as a proxy for
the change in wealth since the last time the
PSID collected wealth d ata, a nding that in-
come is signi cantly related to the probability
of ling for bankruptcy could support either
theory.
12
In our base-case speci cation, we include a s
regressors househ old
i
’s income in year
t 2
1
and t he reduction in household
i
’s income be-
tween year
t 2
2 and year
t 2
1 if income fell,
or else zero. We also esti mate a version of our
model that excludes the income variables, but
includes direct measures of whether adverse
events occurred.
C.
Local Trends
We also test whether households’ bankruptcy
ling decisions are in uenced by the aggregate
bankruptcy ling rates in their localities in the
previous year. As discussed above, there are
differences in the way bankruptcy law is admin-
istered a nd practiced across bankruptcy districts
which make ling persistently more attractive
in certain districts. Because we include state
xed effects in our regressions, p ersistent dif-
ferences between the district and the national
ling rates will be captured by the state xed
effects, except to the extent that districts’ ling
rates differ from their states’ ling rates. How-
ever, an increase in a district’s ling rate may
also start an information cascade which causes
the trend of bankruptcy lings in the district to
differ from the national trend. A survey of re-
cent bankruptcy lers by Visa U.S.A., Inc.
(1997) found that half of them rst heard about
bankruptcy from friends or rel atives. Also, re-
spondents reported t hat they were very appre-
hensive about ling for bankruptcy beforehand,
but fo und the actual process of ling much
quicker and easier than they expected. If house-
holds live in a district with a higher bankruptcy
ling rate, then they a re more likely to hear
rsthand about bankruptcy from friends or rel-
atives because the latter are more likely to have
led. Their friends/relatives will probably te ll
them that ling for bankruptcy is quick and
easy. This information will tend to ma ke house-
holds more comfortable with the idea of bank-
ruptcy, so that the level of bankruptcy stigma
falls and individual households’ probabilities of
ling rise. Higher ling rates then continue the
process of shifting attitudes toward a more fa-
vorable view of bankruptcy.
We test for local trends i n the bankruptcy
ling rate by entering the aggregate ling rate in
the household’s bankruptcy district the previous
year.
13
Because we also include state and year
xed effec ts in our regressions, the coef cient
of the lagged aggregate bankruptcy ling rate
tests whether households are more likely to le
for bankruptcy if they live in districts with
higher aggregate ling rates, controlling for per-
sistent differences across states in bankruptcy
ling rates and for the national trend in bank-
ruptcy ling rates. A signi cant coef cient on
the lagged bankruptcy ling rate in the di strict
could re ect local differences in th e level of
bankruptcy stigma or local differences in the
administration of bankruptcy law that make the
district differ from the s tate, or could re ect the
in ue nce of information cascades.
D.
Other Variables
We include a vector of demographic vari-
ables which may b e related to households’ de-
cisions to le for bankruptcy. These are the age
and age squared of the household head, the
head’s education level, family size, whether the
household owns its o wn home, and whether t he
household head or spouse owns a business. The
12
Hurst et al. (1998) show that over 35 percent of
ve-year wealth changes in the PSID can be explained by
household income, age, education, race, and initial wealth.
13
Bankruptcy court districts are the same as federal
district court districts. There are 90 individual bankruptcy
court districts, with one to four districts in each state. We
are grateful to Ted Eisenberg for providing us with a pro-
gram which assigns counties to federal court districts.
710 THE AMERICAN ECONOMIC REVIEW JUNE 2002
