Who Files for Bankruptcy - Rutgers Business School

Who Files for Bankruptcy?
State Laws and the Characteristics of Bankrupt Households
Michelle M. Miller
Assistant Professor
Rutgers Business School
Department of Finance and Economics
1 Washington Park, Room 1154
Newark, New Jersey 07102
[email protected]
(973) 353-5340
June 2011
Abstract
The characteristics of bankrupt households (such as income levels and asset levels) vary
widely across states. This paper asks whether these variations can be attributed to state
bankruptcy statutes such as property exemption laws or garnishment laws. Using a new
household-level dataset, I find that high exemption levels encourage high asset
households to file for bankruptcy while high garnishment rates encourage low income
households to file. The results support a theoretical model in which households choose
between repayment, bankruptcy, and non-response (which occurs when households
simply “walk away” from their bills). Although previous theoretical models have
ignored non-response, it is, in fact, an important alternative form of “relief” for many
households in financial distress.

I would like to thank Robert A. Margo, Lars J. Lefgren, Randall P. Ellis, and Erik Hurst for their helpful
feedback. In addition, I would like to thank Lance E. Miller, Esq., for his legal expertise. Finally, I am
grateful to the seminar participants at Boston University, Rutgers Business School and the St. Louis Federal
Reserve Bank. All remaining errors are my own.
1. Introduction
Over 800,000 American households filed for bankruptcy in 2007. However, these
petitions, more than 2,000 a day, were not distributed uniformly across the nation. In
Maine, for example, 4.2 of every 1,000 households filed for bankruptcy in 2007
compared to 16.6 per 1,000 households in Tennessee.
Like filing rates, the
characteristics of bankrupt households differed across states. In Maine, households filing
for bankruptcy in 2007 had a mean income of $41,463 whereas, in Tennessee, the
comparable figure was $23,027. This paper explores the sources of these differences,
both theoretically and empirically.
Article I, Section 8 of the United States Constitution specifies that bankruptcy
falls under federal jurisdiction. Accordingly, the basic features of bankruptcy law are
uniform across the county. However, states have adopted a variety of statutes that, in
principle, may affect a household’s bankruptcy decision.
Key examples include:
exemption laws, which protect property from seizure, and garnishment laws, which limit
the extent to which a person’s earnings can be “garnished” by creditors. I demonstrate
that these laws have a heterogeneous impact on a household’s bankruptcy decision--they affect who files for bankruptcy.
To study the impact of state bankruptcy laws, I develop a theoretical model in
which households choose between three options--- repayment, bankruptcy, and nonresponse. Non-response occurs when households simply “walk away” from their bills;
such households may be subject to both asset seizure and wage garnishment (the severity
of both asset seizure and wage garnishment is governed by state law). While Dawsey and
Ausubel (2004) discuss non-response in great detail, it has never been explicitly modeled.
Indeed, previous theoretical models have viewed bankruptcy as the consumer’s only
alternative to repayment (Gropp, Scholz and White 1997; Nelson 1999; Adler, Polak and
Schwartz 2000; Wang and White 2000; Athreya 2002; White 2005; Li and Sarte 2006;
Pavan 2008). However, it is, in fact, necessary to include non-response in any theoretical
model, as it provides an alternative form of “relief” for many households in financial
distress.
To date, because household-level data are largely unavailable, it has been difficult
to estimate the impact of state laws on bankruptcy. Instead, studies relied on aggregate
data, asking whether state laws can explain cross-state variations in filing rates (Apilado,
Dauten and Smith 1978; Shiers and Williamson 1987; Weiss, Bhandari and Robins 2001;
Fisher 2001).1 However, several problems arise when using aggregate data. First, a
cross-state analysis of bankruptcy rates has limited degrees of freedom--- controlling for
the host of demographic and legal factors affecting bankruptcy consumes nearly every
degree of freedom. State-level panel data are of little help because the laws under
analysis rarely change. To better understand the impact of state laws on bankruptcy, I
create a new household-level dataset. This dataset has several advantages. First, with
rich household-level data, I am able to control for many of the potential correlates of
bankruptcy. In particular, the dataset includes detailed information on recent changes to
household income, which I show has a strong influence on the probability of filing for
bankruptcy. By design, aggregate data are not able to control for this factor, and, as such,
are flawed by omitted variable bias. Additionally, by using household-level data, I can
examine the heterogeneous impact of state laws across households; by comparison
aggregate data can only estimate the impact of state laws on the average household.
By examining the impact across households, I can explain two empirical puzzles.
First, to date, there has been no compelling empirical evidence that exemption laws
impact bankruptcy--- this is because the majority of households at risk of bankruptcy
have few assets. As shown in this paper, low asset households are not dramatically
affected by exemption laws. Exemption laws, however, are an important consideration
for households with high asset levels. Additionally, this paper explains why bankruptcy
is more popular among middle income rather than among lower income households. I
show that low income households prefer non-response to bankruptcy; these households
prefer to forfeit portions of their wages and assets in order to avoid the high costs
associated with bankruptcy.
As mentioned, for my empirical analysis, I construct a new household-level
dataset--- it contains information on all households who filed for bankruptcy on February
1
Alternatively, some studies asked whether state laws contribute to district-level (Buckley and Brinig
1998), county-level (White 1987) or zipcode-level (Lefgren and McIntyre 2009) variations in filing rates.
2
1, 2007. For each household, the dataset includes a detailed income history, as well as
asset, debt, and demographic information. A control sample of non-bankrupt households
is constructed using data from the Panel Study of Income Dynamics (PSID). I employ a
choice-based sampling technique to ensure unbiased and consistent estimates and to
correct for the problems of the endogenous sample. I find that high exemption levels
encourage high asset households to file for bankruptcy. I also find that high garnishment
rates encourage low income households to file for bankruptcy. My results are robust to a
variety of specification tests.
2. Institutional Background
Under Article I, Section 8 of the United States Constitution, Congress has the
authority to “establish uniform laws on the subject of bankruptcy throughout the United
States.” Although bankruptcy is governed by federal law, states have enacted a variety of
statutes which are thought to influence a household’s bankruptcy decision. This section
begins with a discussion of state exemption and garnishment laws, detailing the dramatic
variations across states. Then it continues with a discussion of federal bankruptcy law,
noting how each type of state law may impact a household’s bankruptcy decision.
2.1 State Exemption Laws
Exemption laws protect an individual’s unsecured assets both inside and outside
of bankruptcy. There are two types of exemption laws: personal exemption laws and
homestead exemption laws. Personal exemptions allow individuals to retain personal
property (such as jewelry and appliances) while homestead exemption laws protect a
household’s residence. In this paper I concentrate on homestead exemption laws, as they
are substantially larger and more readily quantified.2
2
These exemption levels vary
Personal property exemptions list the type of personal property an individual can protect in bankruptcy.
Typically, they refer to categories of basic necessities, like food, clothes, furnishings or tools of trade, but
sometimes they refer to specific items, like herds of sheep or military uniforms. However, unlike
homestead exemption laws, most personal exemption laws do not place a dollar value on protected assets;
as a result, they are difficult to quantify. In Delaware for example, a household’s clothing, jewelry, books,
family portraits, piano, leased organs, sewing machines, burial plot, and church pew are all protected by
personal exemption laws. Similarly, in Texas, personal exemption laws protect the family Bible, household
pets, two horses with saddle, blanket and bridle for each, twelve head of cattle, 120 fowl, and any food on
hand for these animals. Using several different calculation techniques, Hynes, Malani and Posner (2003)
determined that the average personal exemption law protects approximately $6,000 of assets. By
3
significantly across states--- five states (Florida, Iowa, Kansas, Oklahoma and Texas)
have “unlimited” homestead exemptions meaning that (literally) an unlimited amount of
home equity is protected from seizure by creditors. On the other hand, three states
(Maryland, New Jersey and Pennsylvania) have no homestead exemptions; in these
states, creditors can seize all home equity. It should be noted that exemption laws protect
households against involuntary liens or unsecured debts. However, they do not protect
households against voluntary liens or consensual secured debts. Thus, exemption laws do
not protect households against foreclosure.
Property exemption laws have a long history in the United States. The first
homestead exemption law was passed in Texas (then the Republic of Texas) in 1839.
Coupled with free land, Texas hoped homestead exemptions would attract American
settlers (Hynes, Malani and Posner 2003).3 By 1860, nearly every state in the nation had
adopted a homestead exemption law. And the regional patterns which existed in the
1860s still persist today. Hynes, Malani, and Posner (2003) demonstrate that the best
predictor of a state’s current exemption level is its historical exemption level. Indeed,
many of the current laws still contain extremely archaic provisions.4
2.2 State Garnishment Laws
When a household fails to repay its debts, creditors may garnish the household’s
income. Garnishment can be taken for any type of debt--- common examples include
defaulted child support, taxes, court fines, and student loans.
Wages, salaries,
commissions, bonuses, and income from retirement programs can all be garnished.
Garnishment procedures first appeared in Maryland’s State Statue in 1683. Soon
other states and territories passed similar rulings. In the 1960s, concern arose regarding
garnishment abuses. Therefore, in 1968, Congress enacted the Federal Consumer Credit
Protection Act (CCPA). According to this Act, 75 percent of wages or 30 times the
comparison, the average homestead exemption law protects approximately $50,000 of assets. Given their
size, I am generally not concerned about omitted variable bias.
3
Immigration guides specifically advertised a generous homestead exemption, good soil, and a climate
suitable for cotton.
4
In Oklahoma, for example, a debtor can exempt a gun, twenty head of sheep, and “all provisions and
forage on hand; see 31 Okl. St. § 1 (2000).
4
federal minimum wage per week (whichever is higher) is protected from garnishment.5
In addition, the CCPA prohibits employers from firing a worker due to garnishment. The
CCPA also allows states to enact their own garnishment laws, provided that these laws
protect a greater portion of borrowers’ wages than the federal share. Four states--Florida, Pennsylvania, South Carolina, and Texas--- currently prohibit wage garnishment
except for debts related to taxes, child support, federally guaranteed student loans, courtordered fines or restitution for a crime the debtor committed. An additional 22 states
have thresholds that are higher than the federal law.
2.3 Personal Bankruptcy in the United States
The number of American households seeking bankruptcy relief has increased
substantially over the past three decades. While less than 250,000 filed for bankruptcy in
1978, over 2 million households filed for bankruptcy in 2005. In response, Congress
enacted the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005
(BAPCPA).6
For households facing serious debt, there are many advantages to filing for
bankruptcy. First, through the bankruptcy process, households are able to discharge
(legally default on) much of their debt--- the average household discharges approximately
$36,000 of debt upon filing for bankruptcy (Culhane and White 1999). In addition, once
the household files for bankruptcy, all wage garnishment must stop. In states where a
sizeable portion of wages are garnished, this is clearly beneficial to the household’s cash
5
On February 1, 2007, the federal minimum wage was $5.15. Thus, at the time my sample was collected,
at least $154.50 (30 x $5.15) of weekly wages was exempt from garnishment. When an employee’s weekly
wages exceeded $154.50 but were less than $206.00, only the amount over $154.50 could be garnished.
For example, if an employee earned $165 in a particular week, only $5.50 could be garnished. When an
employee’s earnings were $206.00 or more in a given week, up to 25 percent of those earnings could be
garnished. For an employee earning $250.00 a week, 25 percent of his earnings (or $62.50) could be
garnished while $187.50 had to be paid to the employee.
6
The BAPCPA made several amendments to the Bankruptcy Code in hopes of curbing abusive filings. For
example, the BAPCPA hoped to decrease the number of repeat filers. However, Miller and Miller (2008)
found that the law had virtually no impact on the rate of repeat filings. The authors find that the law did not
appear to affect who repeatedly filed for bankruptcy; the financial description of repeat filers remained the
same before and after the BAPCPA. Instead, the BAPCPA seems to have increased the time debtors
waited between filings. The BAPCPA also hoped to prevent high income households from filing under
Chapter 7 of the Bankruptcy Code. In particular, the “means test” required that households earning income
above the state median level file under Chapter 13. However, most papers have found this income
restriction rarely applies (Flynn and Bermant, 2000; Culhane and White, 1999; Tabb and McClelland,
2007; White, 2007).
5
flow.
However, in states where only a small portion of wages can be garnished,
bankruptcy may be less advantageous. Regardless of the financial benefits, bankruptcy is
a costly endeavor. In addition to the nearly $300 filing fee, households must pay attorney
fees, which average $1,830 (Palank 2008). Moreover, there are non-pecuniary costs to
bankruptcy, including the stigma of bankruptcy and future restrictions from the credit
market. And finally, households must either forfeit a portion of their assets or a portion
of their future income.
The United States has two primary procedures for personal bankruptcy--- Chapter
7 and Chapter 13. Chapter 7 is intended for households with little or no income while
Chapter 13 is designed for households with regular income.
2.3.1 Chapter 7: Liquidation
Under Chapter 7 of the Bankruptcy Code, households must liquidate all of their
nonexempt assets; these are the assets above the exemption levels discussed above.7 As
mentioned, five states have unlimited homestead exemption levels; in these states debtors
will not be forced to sell their house to payoff their creditors. However, in the three
states with no exemption level, households must surrender their home.
Clearly,
bankruptcy is more advantageous in states with higher exemption levels.
After nonexempt assets are liquidated, the proceeds are distributed amongst the
household’s creditors. The case is then closed and the household’s remaining debts are
discharged. Most unsecured debts, including credit card debts, installment loans, medical
debts, unpaid rent and utility bills, tort judgments, and business debts, can be discharged
under Chapter 7. Under Chapter 7 of the Bankruptcy Code, the debtor may keep all of
his future earnings.
However, not all households can file for Chapter 7 relief.
Households who filed for Chapter 7 within the past seven years are ineligible.8
7
In some states a debtor is required to file a declaration of homestead exemption for such exemption to be
enforceable. However, the household can file this declaration post-petition, so as a practical matter, this
requirement does not often interfere with a debtor's ability to claim a homestead exemption. Accordingly,
this paper does not make a distinction between states which automatically provide homestead exemptions,
and those which require the filing of a homestead declaration (In re Michael, 163 F.3d 526 9th Cir. 1998).
8
Specifically, the Bankruptcy Code states that debtors are only eligible for a Chapter 7 discharge every
eight years.
6
Additionally, the BAPCPA precludes debtors with income above their state’s median
income from filing under Chapter 7.
2.3.2 Chapter 13: Debt Reorganization
Under Chapter 13, households retain all of their assets and instead agree to repay
some of their debts from future earnings. Debtors pay their projected monthly disposable
income (the difference between their monthly income and monthly budgeted living
expenses) into the Chapter 13 repayment plan.9 After sixty months, the case is closed,
and any remaining debts are discharged. It is important to note that the repayment plan
must compensate creditors at least as much as they would receive under Chapter 7. All
debtors are permitted to file under Chapter 13 of the Bankruptcy Code. Debtors can file
for Chapter 13 bankruptcy every two years.10
In 2007, less than 40 percent of cases were filed under Chapter 13 of the
Bankruptcy Code. The major benefit of filing a Chapter 13 is asset retention.
In
addition, a Chapter 13 bankruptcy only remains on a debtor’s credit report for seven
years; by comparison, a Chapter 7 bankruptcy remains on a debtor’s credit report for ten
years. Thus, there is reason to believe that Chapter 13 filers have better access to future
credit. But, compared to Chapter 7, Chapter 13 is an extensive and lengthy process--under Chapter 13, households must make monthly payments for five years.11 In contrast,
Chapter 7 cases typically last less than six months.12 As shown in Lefgren, McIntyre and
Miller (2010), the best observable predictor of chapter choice is the consumer’s attorney;
households are likely to file under Chapter 13 if they consult a bankruptcy lawyer who
specializes in Chapter 13 cases.
To conclude this section, Table 1 provides a summary of state exemption and
garnishment laws; with this table, the cross-state variation in these laws becomes
9
The Chapter 13 Trustee distributes these funds between the household’s creditors.
Specifically, if a debtor has previously filed for under Chapter 13 of the Bankruptcy Code, he is eligible
for another discharge after two years. And if the debtor has previously filed under Chapter 7 of the
Bankruptcy Code, he is eligible for a Chapter 13 discharge after four years.
11
If a household fails to make the court mandated payments the case is dismissed and the household is once
again liable for all its debts. Lefgren, McIntyre and Miller (2008) report that 60 percent of bankruptcies
filed under Chapter 13 of the Bankruptcy Code are dismissed largely due to nonpayment on the debtor’s
part.
12
In fact, if a household does not have any non-exempt assets, the case will typically only last one month.
10
7
apparent. In terms of homestead exemptions, while five states have unlimited homestead
exemptions, three states have none. And although 24 states allow garnishment up to the
federal limit, 26 states have set thresholds higher than the federal law, including four
states which prohibit wage garnishment altogether. 13
3. Literature Review
To date, it has been difficult to study the impact of state laws on a household’s
bankruptcy decision. As household-level data were largely unavailable, early works used
aggregate data, regressing state or district filing rates on legal variables. However,
several empirical concerns arise when using aggregate data. First, a cross-state or crossdistrict analysis of bankruptcy rates has limited degrees of freedom---controlling for the
host of demographic and legal factors affecting bankruptcy consumes nearly every degree
of freedom.14 Using state-level panel data (Buckley and Brinig 1998; Fisher 2001)
cannot solve this problem as the state laws under analysis rarely change.15 A more
important criticism, however, is the inability of aggregate data to estimate the
heterogeneous impact of state laws across households--- aggregate data only capture the
impact of state laws on the average consumer. Indeed, state laws have a varied impact
across households, which can only be identified with household-level data.
Recently, as datasets have become available, several papers have used householdlevel data to examine the effects of state laws on bankruptcy. None of these papers,
however, have studied the heterogeneous impact across households. For example, using
individual-level data from a credit card issuer, Dawsey and Ausubel (2004) found that
high homestead levels and high garnishment rates encourage bankruptcy. However, their
regressions only included level, not interaction, terms. Thus, the authors have once
again, assumed that state laws have a uniform impact on all households. Also using
household-level data, Fay, Hurst and White (2002) examined whether the likelihood of
13
As shown in Lefgren and McIntyre (2009) these state laws are not correlated. It is not the case that a
state protects households from their creditors by passing both high homestead exemption laws and low
garnishment rates. Indeed the correlation between these two laws is 0.00.
14
One of the most convincing papers to date, Lefgren and McIntyre (2009), conserves degrees of freedom
by first regressing zip code bankruptcy rates on demographic variables and then using the results to
construct indices of states’ demographic propensity to file for bankruptcy.
15
Panel methods can only identify short-run responses to policy changes, which could be quite different
from the long-run effects.
8
filing depends on the “financial benefit” of bankruptcy. Financial benefit is defined as
the debt that can be discharged less nonexempt assets a debtor loses by filing. However,
with this definition, the impact of exemption levels cannot be disentangled from
additional units of assets or debts.
In summary, previous studies have been powerless in exploring the heterogeneous
impact of state laws across households. Early works used aggregate data, and therefore,
could only estimate the impact of laws on the average consumer. Later studies, which
used household-level data, were also limited; either these studies assumed that laws had a
homogenous impact on households, or the impact of laws could not be disentangled from
the impact of other factors.
By examining the impact across households, I can explain two empirical puzzles.
First, there has been no empirical evidence that exemption laws impact bankruptcy. 16 As
noted by Lefgren and McIntyre (2009), this is because the majority of households at risk
of bankruptcy have few assets; as I show in this paper, low asset households are not
dramatically affected by exemption laws. However, by including interaction terms, I
show that exemption laws are important to households with high levels of assets.
Additionally, this paper explains why bankruptcy is more popular among middle rather
than among lower income households. While conventional wisdom has supported this
claim, economic theory has been unable to explain this result. I show that low income
households prefer non-response to bankruptcy; these households prefer to forfeit portions
of their wages and assets in order to avoid the high costs associated with bankruptcy.
4. Model
In this section, I develop a two-period model of bankruptcy. The purpose of this
model is to show how state exemption and garnishment rates affect a household’s
bankruptcy decision. In the first period, households learn their first period income, and
16
While White (1987) found that high exemption levels had a positive impact on bankruptcy rates, virtually
all other published papers found either no statistically significant effect (Peterson and Aoki 1984; Weiss,
Bhandari and Robins 2001) or a negative effect (Apilado, Dauten and Smith 1978; Shiers and Williamson
1987; Buckley and Brining 1998).
9
choose their consumption, assets, and debts.17 In the second period, armed with their
assets and debts, households learn their second period income, and decide whether to file
for bankruptcy. Risk neutral lenders are assumed to make zero profits.
In order to be consistent with bankruptcy law, assets represent home equity while
debt represents unsecured loans such as credit card debts.
This model does not
distinguishing between the value of the home and the amount of the mortgage; in order to
make this distinction, the model would need to include an additional choice variable.
Again, the purpose of this model is to show how state exemption levels impact a
household’s bankruptcy decision. As exemption levels only apply to non-consensual
secured debts, this simplification should not be worrisome.18 Extending the model to
include the value of the home and the amount of the mortgage is left to future works.
This model makes two extensions to previous works.
First, it expands the
traditional one-period bankruptcy model to a two-period model. Thus, the model allows
assets, debts, and the interest rate to depend on the state’s exemption or garnishment
laws; earlier works have assumed that these variables are exogenous. 19 Additionally, in
this model, households choose between three options--- repayment, bankruptcy and nonresponse. Previous models have viewed bankruptcy as the consumer’s only alternative to
repayment. However, households can simply ignore collection letters and bills--- such
behavior is deemed non-responsive. Non-responsive households are subject to both wage
garnishment and asset seizure. While Dawsey and Ausubel (2004) discuss non-response,
it has never been explicitly modeled.
Non-response is an important alternative to
consider. In 2007, when this data for this paper was collected, 7.1 percent of U.S.
families reported to be at least 60 days behind on one of their loans (Bucks et al. 2009).
17
Without bankruptcy, a consumer will either borrow or save; he will not do both. However, when
bankruptcy is available, households may borrow and save simultaneously. Although debt requires paying
an interest rate premium, the prospect of not having to repay the debt in full lowers this premium.
18
This model does not address consensual secured loans. Again, the purpose of this model is to show how
state exemption levels impact a household’s bankruptcy decision. Fundamentally, homestead exemptions
do not protect a homeowner against foreclosure by a mortgagee. Homestead exemptions apply only against
involuntary liens or unsecured debts. Because exemptions are not relevant to consensual secured debt,
mortgages are not modeled here.
19
Repetto (1998) and Pavan (2008) evaluated the impact of exemption laws on household asset
accumulation. However, neither paper considered the impact of garnishment rates. Furthermore, neither
paper included non-response in their theoretical model.
10
4.1 The Household’s Problem
In this model, risk averse households live for two periods and maximize their
expected lifetime utility. In the first period, households receive income Y1. However,
second period income (Y2) is uncertain; Y2 is uniformly distributed between 0 and 1. In
the first period, households choose their first period consumption (C1), assets (A) which
can be used for future consumption, and debt (D). Note that lenders charge interest rate r
on a loan; as discussed below, I allow the interest rate to depend on the household’s
assets and debts as well as the state exemption level and garnishment rate.20 Thus, the
household’s first period budget constraint is given by C1  Y1  A  D .
In the second period, armed with assets A and debt D(1+r), a consumer learns his
income and decides whether to file for bankruptcy. If he files for bankruptcy under
Chapter 7, the consumer must surrender nonexempt assets and pay a filling fee. All
remaining debts are then discharged (forgiven).
Thus, if the consumer files for
bankruptcy under Chapter 7 his consumption is given by C2 ≤ Y2 + min(A,E) - F where E
represents the state exemption level and F denotes the filing fee.21 If the consumer files
for bankruptcy under Chapter 13 of the Bankruptcy Code, he must surrender his income
and pay a filing fee.
However, the consumer may keep all its assets.
remaining debts are then discharged (forgiven).
Again, all
Thus, the consumer’s consumption
under Chapter 13 of the Bankruptcy Code is given by C2  A  F .22 According to the
Bankruptcy Code, households must repay creditors at least as much in a Chapter 13
repayment plan as in a Chapter 7. In addition, creditors can petition to convert a Chapter
7 into a Chapter 13 if it is in their best interest. Therefore, if the household files for
bankruptcy,
its
consumption
is
given
20
by
C2  min[Y2  min( A, E)  F , A  F ] .
This set-up assumes that assets pay an interest return normalized to unity while debts carry a gross
interest rate of r. This simplification does not alter my model’s predictions.
21
In addition to the filing fee, F incorporates attorney fees, the stigma of bankruptcy, and the cost of future
credit market exclusion.
22
More realistically, when filing under Chapter 13, a consumer surrenders a portion of its income and his
consumption is given by C2≤ βY + A - F. According to the means test, a household can deduct
standardized expenses from its income; standard deductions can be taken for healthcare, housing, utilities,
vehicle expenses, and public transportation expenses. In addition, to the standard deductions, a household
can also deduct actual spending on taxes, payroll deductions, term life insurance, court-ordered payments,
utilities and health care. Therefore, as mentioned above, it may be more realistic to state that when filing
under Chapter 13, a household’s consumption is given by C2≤ βY + A – F. However, the simplification in
the text does not alter my model’s predictions.
11
According to this equation, and in accordance with the Bankruptcy Code, higher income
households file under Chapter 13 rather than Chapter 7.
If a consumer does not file for bankruptcy, his second period consumption is
given by C2  Y2  A  Q(Y2 , A, D, r ) .
Q(Y2,A,D,r)
=
D(1+r).
When
When the household repays its debt in full
the
household
chooses
non-response,
Q(Y2,A,D,r)=gY2+max(A-E,0), where g denotes the state’s garnishment rate. Recall that
when a consumer fails to repay his debts, he is subject to wage garnishment (at rate g)
and creditors may sue the household for his non-exempt assets, max(A-E,0). Thus,
Q(Y2,A,D,r)=min[gY2+ max(A-E,0), D(1+r)].23
Therefore in the second period, households file for bankruptcy when
min[Y2  min( A, E)  F , A  F ]  Y2  A  Q(Y2 , A, D, r )
or
equivalently,
when
Y2  min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) .
For descriptive purposes, Figure 1 depicts the household’s second period
decisions for a household with sufficiently large assets (A>E) and sufficiently large debts
(D(1+r)>F). The vertical axis depicts second period consumption while the horizontal
axis depicts all the possible realizations of second period income. Consumption can be
divided into three regions. In the left-most region, where Y2 
F
, households are nong
responsive--- they do not repay their debt and, as a result, wage garnishment and asset
seizure take place. These households avoid bankruptcy (and its high filing fees) without
repaying their debts in full.
In the middle region (where
households file for bankruptcy. In particular, households with
under
Chapter
7
of
the
Bankruptcy
Code
F
 Y2  D(1  r )  F )
g
F
 Y2  A  E will file
g
while
households
with
A  E  Y2  D(1  r )  F will file under Chapter 13. Finally, in the right-most region
(where Y2  D(1  r )  F ) households repay their debts in full. As seen clearly in this
23
Households who choose non-response may also suffer from non-pecuniary repercussions such as stigma
or future exclusion from the credit market. One could write Q(Y2,A,D,r)=min[gY2+ max(A-E,0)-f, D(1+r)]
where f incorporates the stigma of non-response and the cost of future credit market exclusion. Because f <
F, this simplification does not alter the model’s predictions.
12
figure, the model predicts that bankruptcy is primarily used by middle income
households.
With the second period decisions detailed above, the household’s maximization
problems can be written as: 24, 25
max( A E , 0 )
F/g
max U (Y1  A  D) 
A, D
 U ((1  g )Y
2
 U (Y
 min( A, E ))dY2 
2
0
D (1 r )  F
 U ( A  F )dY
2
max( A E , 0 )
 min( A, E )  F )dY2
F/g
1

 U (Y
2
 A  D(1  r ))dY2
D (1 r )  F
The household’s first order condition with respect to debt is as follows:26
U (Y1  A  D)  (1  r  D
1
dr
)
U (Y2  A  D(1  r ))dY2  0 .
dD D (1r ) F
The first term of this equation denotes the first period utility gain associated with an
additional unit of debt. An additional unit of debt only affects second period utility if the
household chooses repayment; the second term of this equation captures the second
period utility loss associated with additional units of debt, integrated over the region of
repayment.
The first order condition with respect to assets is:
F/g
 U (Y1  A  D) 

U ((1  g )Y2  A)dY2 *
0
d min( A, E )

dA
 [ D(1  r )  F  max( A  E ,0)]U ( A  F )  (1  D
max( A  E , 0 )

F/g
U ' (Y2  A  F )dY2 *
d min( A, E )
dA
1
dr
)
U (Y2  A  D(1  r ))dY2  0
dA D (1r )  F
The first term of this equation represents the first period utility loss associated with an
additional unit of assets.
An additional asset increases second period utility if the
24
I assume that the household’s utility is only defined over their consumption. For an alternative model see
Livshits, MacGee and Tertilt (2007).
25
It is important to note that households do not choose their exemption level E. According to Section
548(b)(3)(A) of the Bankruptcy Code, debtors cannot manipulate homestead exemptions by moving to
another state. This section states that if a debtor moved within the past 730 days, he is bound by the
exemption level of the state in which he previously resided.
 ( x)
 ( x)
26
dG is used to find the first order
Leibnitz rule d G( x, t )dt G( x,  ( x)) d  G( x,  ( x)) d 
dt


dx  ( x )
dx
dx  ( x ) dx
conditions.
13
household chooses non-response (and has assets below the exemption level), bankruptcy
under Chapter 7 (and has assets below the exemption level), bankruptcy under Chapter
13, or repayment.
The second term denotes the second period utility gain if the
household chooses non-response, when the consumer does not have non-exempt assets,
integrated over the region of non-response. The third term captures the marginal utility
of assets when the consumer files for bankruptcy under Chapter 7, when the consumer
does not have non-exempt assets, integrated over the region of Chapter 7. The fourth
term equals the marginal utility of assets when the household files for bankruptcy under
Chapter 13, multiplied by the probability that a household files for bankruptcy under
Chapter 13.
Finally, the last term denotes the marginal utility of assets when the
household chooses repayment, integrated over the region of repayment.
4.2 The Lender’s Problem
Lenders are assumed to be risk neutral. Their zero profit condition is given by:
max( A E , 0 )
F/g
 [ gY
2
 max( A  E ,0)]dY2 
0
 max( A  E,0)dY
2
F/g
D (1 r )  F

 Y dY
2
max( A E , 0 )
2
1

 D(1  r )dY
2
 D(1  r f )
D (1 r )  F
where the first term represents partial repayment from non-response, the second term
represents partial repayment under Chapter 7, the third term represents partial repayment
under Chapter 13, the fourth term represents full repayment and rf is the risk free rate of
return. If no interest rate satisfies this equation, lenders will not lend. I assume that
lending is limited by D(1+r) < 1.
The lender’s first order condition with respect to the interest rate is:
max( A  E,0) *
d max( A  E,0)
dD
 [ D  (1  r )
] * [1  D(1  r )]  0
dr
dr
The first term represents the lender’s gain if the household files under Chapter 7 of the
Bankruptcy Code and has non-exempt assets. Recall that when a household files under
Chapter 7 of the Bankruptcy Code, it must use its non-exempt assets to repay its debts. If
a lender increases the interest rate, households will increase their assets; thus, when a
household files for bankruptcy under Chapter 7, the lender will receive a higher payment.
14
The second term represents the lender’s gain if the household repays its debts in full.
This term represents several forces. If a lender increases the interest rate, households will
decrease their debts. In addition, households are less likely to repay in full.
4.3 Predictions about Assets, Debts and the Interest Rate
The first order conditions are totally differentiated and Cramer’s rule is used to
solve for dr/dE ≥ 0.27 High exemption levels protect assets should the household choose
non-response or bankruptcy. Thus, lenders collect less when households choose nonresponse or if they file for bankruptcy. As a result, lenders will charge a higher interest
rate in states with high exemption levels.
Similarly, after totally differentiating the first order conditions and applying
Cramer’s rule, the model predicts that dA/dE ≥ 0 and dD/dE ≥ 0.
Thus, in high
exemption states, households will simultaneously accumulate assets and debts; Lehnert
and Maki (2002) refer to this behavior as “borrowing to save.” High exemption levels
protect assets should the household choose non-response or bankruptcy. Thus, under
either of these options, households can retain a larger amount of assets without having to
repay their debts. This creates an incentive for households to accumulate additional
assets at the cost of additional units of debts.
The model predicts similar behavior from households in high garnishment states
(dA/dg ≥ 0 and dD/dg ≥ 0). A household is more likely to file for bankruptcy in a high
garnishment state. And because it is preferable to have high assets and high debts when
filing, households in high garnishment states will simultaneously accumulate both assets
and debts.
Furthermore, the model predicts that dr/dg ≥ 0. In states with high garnishment
rates, lenders can garnish a larger portion of the household’s income. This creates two
opposing effects. Because lenders collect more when a household chooses non-response,
high garnishment rates create an incentive for lenders to lower interest rates. On the
other hand, more households will file for bankruptcy (instead of choosing non-response);
these households will repay less debt. This creates an incentive for lenders to increase
27
For more details, see the Mathematical Appendix.
15
interest rates. This second effect is larger, and, as a result, the interest rate is higher in
states with high garnishment rates.
4.4 Predictions About Bankruptcy
Again,
households
file
for
min[Y2  min( A, E)  F , A  F ]  Y2  A  Q(Y2 , A, D, r )
Y2  min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) .
bankruptcy
or
equivalently,
when
when
From this condition, the model
makes several predictions about the likelihood of bankruptcy.
Predictions about the Likelihood of Bankruptcy:28

High exemption levels encourage high asset households to file for bankruptcy.

High garnishment rates encourage low income households to file for bankruptcy.

The likelihood of bankruptcy is decreasing in A.

The likelihood of bankruptcy is increasing in D.
5. Data Description
To test the theoretical model detailed above, I use a choice-based sampling
technique. 29 As data on bankrupt households are difficult to obtain, empirical studies on
personal bankruptcy often rely on choice-based sampling. For example, Domowitz and
Sartain (1999) match a sample of households who filed for bankruptcy in 1980 with
households from the 1983 Survey of Consumer Finances (SCF).
Similarly, Zhu
(forthcoming) augments bankruptcy filing data from 2003 with detailed household and
consumption data from the 2004 SCF. The dataset used in this paper comes from two
distinct sources: information on bankrupt households was hand-collected from
bankruptcy petitions while data on non-bankrupt households came from the Panel Study
of Income Dynamics (PSID).
It should be noted that the ideal dataset would combine three samples of
households: households who chose non-response, households who chose bankruptcy, and
28
For more details, see the Mathematical Appendix.
Choice-based sampling arises when selection into the sample is determined by the dependent variable.
Manski and Lerman (1977), as discussed in more detail below, show that choice based sampling techniques
can correct for the problems of the endogenous sample and ensure consistent and unbiased estimates.
29
16
households who chose repayment. Unfortunately, a large dataset on non-responsive
households is not readily available. In Section 7.4, I create a smaller alternative choicebased dataset, in which households chose between these three options. My findings are
robust to this alternative specification.
5.1 Bankruptcy Sample
When filing for bankruptcy, debtors must complete a bankruptcy petition: a
detailed form describing their finances.30 This petition is then archived on PACER
(Public Access to Court Electronic Records), the court’s centralized registration and
billing website. With a subscription to this website, I created a unique dataset of the
1,694 households that filed for bankruptcy on February 1, 2007. 31,32
Bankruptcy petitions provide an abundance of financial information.
First,
households itemize their unsecured debts including credit card debts and medical bills.33
In addition, households must record the value of their home as well as any secured claims
on their home. Thus I can calculate the household’s home equity; because exemption
laws protect equity in one’s home equity, I used home equity as my measure of assets. In
addition, debtors must provide comprehensive income information--- they detail their
income for the past two calendar years. Finally, court records provide the debtor’s
address, marital status, and household size. Other demographic information, including
age and education are unfortunately, unavailable. However, since I know the filer’s
30
Households provide information under penalty of perjury. In addition, the household’s attorney is
required to verify all financial statements.
31
As the petitions are only available in a PDF format, the relevant information then had to be extracted into
a more usable format. This dataset is a nationwide census of households who filed for bankruptcy on
February 1, 2007.
32
There are several advantages to collecting data from February 1, 2007. Because my data was collected in
2007, my sample consists of data from cases filed after the BAPCPA. This paper is one of the first to
examine bankrupt households since the Bankruptcy Code was amended in 2005. Additionally, because I
collected data from 2007, I was able to could collect a nationally representative dataset--- prior to 2007
many bankruptcy courts did not use PACER’s Case Management/Electronic Case Files (CM/ECF) system.
Therefore, to collect a nationally representative sample I would have had to travel to each bankruptcy court.
February 1st was selected as a random date at the beginning of the year; the bankruptcy petitions collect
information on the household’s income from the previous two calendar years. Thus, for petitions filed on
this date, I likely have accurate income information.
33
In particular, from Schedule F, I extract information on a household’s unsecured non-priority debts. In
general, these are the debts which households can discharge in bankruptcy. My results are statistically
similar when I include unsecured priority debts (from Schedule E). Unsecured priority debts include
domestic support obligations and unpaid taxes.
17
address, I use 2000 Census block statistics to estimate these demographic
characteristics.34
5.2 Control Sample
I augment the choice-based sample of bankrupt households with households from
the 2007 PSID. Because bankrupt households are not identified in this study, I begin by
assuming that none of the households in the PSID filed for bankruptcy. This assumption
is tested in Section 7.3.
Each of the 6,996 households in the PSID details its income from the previous
two calendar years. In addition to annual measures of family income, the PSID provides
a detailed inventory of the family’s home equity and unsecured debts. Demographic
information is also provided.
5.3 Descriptive Statistics
Summary statistics, comparing bankrupt and non-bankrupt households, can be
found in Table 2. Several facts are worth noting. First, these statistics confirm the
conventional notion that bankrupt households have lower incomes. In the first year,
bankrupt households earned nearly $20,000 less than non-bankrupt households. And in
the second year, bankrupt households earned over $37,000 less than non-bankrupt
households. Not surprisingly, bankrupt households are also characterized by lower levels
of home equity ($9,833 compared to $95,416). And, bankrupt households have higher
levels of debt; while households in the PSID have an average $8,017 of unsecured debt
bankrupt households have an average $50,876 of unsecured debt.
Known demographic variables including family size and marital status are similar
for households in the bankrupt and supplemental samples. There are two differences
between the samples--- although less likely to be married (and more likely to be single or
separated), bankrupt households are larger. The imputed demographic characteristics,
34
Prior research suggests that missing demographic data should not be a concern. In fact, earlier works
argue that once financial characteristics are controlled for, demographic characteristics are irrelevant. In
interviewing 400 households, Stanley and Girth (1971) found that demographic considerations did not
influence the bankruptcy decision. Domowitz and Eovaldi (1993) confirmed this result; they found that
race was statistically insignificant in explaining aggregate filing rates. As shown in the third column of
Table 3, my results do not change when I exclude the imputed demographic variables.
18
based off the bankrupt household’s census block, indicate that bankrupt households are
younger.
6. Estimation
6.1 Probability of Bankruptcy
In estimating the impact of state laws on a household’s bankruptcy decision, my
primary specification is:35
Prob(Bankrupt) = α0 + α1High Exemption + α2High Exemption*Assets +
α3High Garnishment + α4High Garnishment*Income + α5Assets +
α6Debt + α7X + α8Region + εit
(1)
where Bankrupt is a dummy variable equal to one if the household files for bankruptcy.
High Exemption36 and High Garnishment37 denote high exemption levels and high
garnishment rates respectively.
Assets, Debt, and Income are a household’s assets
(measured by home equity), debts (measured by unsecured debts), and income. X is a
vector of other variables (including income, change in income, and demographic
characteristics). Finally, Region is a vector of regional dummy variables.
Recall, the model predicts that high exemption laws encourage high asset
households to file for bankruptcy (α1 = 0 and α2 > 0) and that high garnishment rates
encourage low income households to file for bankruptcy (α3 > 0 and α4 < 0). Finally, the
model predicts that the likelihood of bankruptcy is decreasing in assets and increasing in
debts (α5 < 0 and α6 > 0).
35
Again, my theoretical model supports a multinomial logit regression as households are allowed to
choose between non-response, bankruptcy and full repayment. However, my dataset does not distinguish
between non-response and full repayment. Therefore my main empirical analysis employs a logit
regression, estimating the probability that a household files for bankruptcy (versus the alternative that they
do not file and choose either non-response or repayment in full). For an alternative specification, using a
smaller dataset, see Section 7.4.
36
A survey of state exemption laws reveals that one third of states have exemption levels greater than the
mean exemption of $50,000. Accordingly, the dummy variable “High Exemption” equals one if the
household lives in a state with a homestead exemption greater than $50,000. As seen in Table 3, my
findings are robust to different definitions.
37
As discussed in Section 2, under federal law, creditors may garnish twenty five percent of wages. Many
states, however, have placed further limits on the garnishment rate. The dummy variable “High
Garnishment” equals one in states that use the federal garnishment limits. As seen in Table 3, my findings
are robust to alternative specifications.
19
If the sample were random, equation (1) could be estimated using a simple logit
regression. However, the sample at hand is not random; instead it is choice-based.38
Therefore, as detailed in Manski and Lerman (1977) equation (1) must be estimated using
weighted exogenous sampling maximum likelihood estimation (WESMLE).
By
controlling for the oversampling of bankrupt households, I am able to obtain consistent
estimates. Specifically, I weight each term by the inverse of the ex-ante probability that
an observation is included in the sample. Let Q1 denote the fraction of the population
that is bankrupt and H1 denote the fraction of the sample that is bankrupt. Then each
bankrupt household is weighted by Q1/ H1 and each non-bankrupt household is weighted
by (1-Q1)/(1-H1).
Regression coefficients and marginal effects are reported in Table 3. All errors
are clustered at the state level. In column I, I estimate the probability of bankruptcy
without any interaction terms. Akin to the large literature detailed earlier, the dummy
variable “High Exemption” does not have a significant impact on bankruptcy. And
consistent with Dawsey and Ausubel (2004), Lefgren and McIntyre (2009) and Fisher
(2001), there is a positive relationship between high garnishment rates and bankruptcy--this result is unsurprising as debtors often declare bankruptcy in order to stop
garnishment.39
In column II, I include the interaction terms suggested by my model--- the results
are consistent with the model’s predictions. First, high exemption levels encourage high
asset households to file for bankruptcy--- although the dummy variable “High
Exemption” is insignificant, it has a positive and statistically significant effect when
interacted with home equity. Prior studies have not found a significant relationship
38
Choice-based sampling arises when selection into the sample is determined by the dependent variable.
Like many surveys, I have oversampled an infrequently made choice. With such a dataset, logit estimates
will be biased and inconsistent (Manski and Lerman 1977). To see the inconsistency of standard binary
choice methods, consider the logit model when the only regressor is the intercept. Then Λ(Xi’β) = Λ(β) and
the logit MLE first-order condition is N-1Σ(Yi - Λ(β))=0. Thus, ˆ =ln( Y /(1- Y )). Consistency of ˆ clearly
requires a random sample; oversampling Y=1 leads to overestimation of Y and hence overestimation of ˆ .
39
Because I am examining behavior after the BAPCPA of 2005 my estimates are smaller than earlier works
(Lefgren and McIntyre, 2009; Dawsey and Ausubel, 2004). The BAPCPA increased filing fees and
required debtors to complete additional paperwork. Because it was cheaper and easier to file for
bankruptcy prior to 2005, households would have been more inclined to file for bankruptcy when faced
with wage garnishment. However, since the BAPCPA, households are less inclined to file.
20
between exemption levels and bankruptcy. This is because most households at risk of
bankruptcy have low levels of assets, and thus, are not dramatically affected by
exemption levels. However, by including an interaction term, I show that exemption
levels have a statistically significant impact on high asset households.
I also find that high garnishment rates encourage low income households to file
for bankruptcy. Because of the high filing fees, many low income households prefer nonresponse to bankruptcy. However, as garnishment rates increase, bankruptcy becomes
more beneficial. High income households on the other hand, prefer to pay the fixed filing
fee rather than suffer from wage garnishment. Therefore, garnishment rates have little
impact on high income households.
As predicted, bankruptcy is primarily used by middle income households; the
coefficients indicate that the probability of bankruptcy is inversely U shaped in income.
Households at the bottom of the income distribution do not file for bankruptcy--- instead
they choose non-response. And households at the top of the income distribution prefer to
repay their debts.
These regressions also show that a negative shock to income increases the
likelihood a household will file for bankruptcy. Rather surprisingly, there has been
relatively little empirical work relating income shocks to bankruptcy.40 One exception,
Fay, Hurst and White (2002), used data from the PSID to examine whether adverse
events increase the likelihood of filing for bankruptcy.
The authors found that a
reduction in income increased the probability of filing for bankruptcy. However, as these
regressions did not control for wealth, the authors worried that their estimates may suffer
from omitted variable bias.
40
Because traditional household-level datasets do not include sufficient income or bankruptcy information,
earlier studies have instead relied on qualitative data. For example, Sullivan et al. (2000) surveyed
households who filed for bankruptcy in 1991. Based on descriptive evidence, they found that most
households filed for bankruptcy after adverse events reduce their income. Nearly two-thirds of
bankruptcies studied cited job-related financial distress as the cause of bankruptcy. Likewise, Repetto
(1998) showed that among a sample of 1996 bankruptcy filers, 20 percent named job loss as the main cause
of their bankruptcy filing decision. While these papers provide important qualitative evidence that adverse
shocks trigger bankruptcy, they cannot decisively link the two events. During an interview, households
may blame bankruptcy filings on adverse events because they are embarrassed, because such events are
more memorable, or because they do not realize their consumption styles jeopardize their financial security
in the first place.
21
As expected, the probability of bankruptcy is increasing in debt.
When a
household files for bankruptcy it can discharge (be freed of) its debts. Accordingly, it is
rather predictable that a household is more likely to file for bankruptcy as its debt rise.
Finally, the likelihood of bankruptcy is decreasing in assets. By liquidating these assets,
households can repay their debts and avoid bankruptcy and its high filing fee.
In columns III-VI, I show that my results are robust to alternative specifications.
In column III, I re-run the regression, using only known demographic variables--- my
results are not sensitive to the exclusion of the imputed demographic variables.
In columns IV-VI, I show that my results are robust to alternative categorical
cutoffs. In column IV, instead of including the variables High Exemption and High
Exemption*Assets, I include a continuous variable denoting the household’s non-exempt
assets (these are the household’s assets above the exemption level). As expected, the
likelihood of bankruptcy is decreasing in non-exempt assets--- again, this result shows
that high exemption levels encourage high asset households to file for bankruptcy. In
column V, I use the state’s actual exemption level (a continuous variable) and its
interaction with home equity.
Again, my results show that high exemption levels
encourage high asset households to file for bankruptcy.
In column VI, instead of
including the variable High Garnishment and High Garnishment*Income, I include a
dummy variable which equals one if any garnishment is allowed in the state. Even with
this alternative specification, it is clear that high garnishment rates encourage low income
households to file for bankruptcy.
In Table 4, to better understand the heterogeneous impact of state laws on a
household’s bankruptcy decision, I examine the likelihood of bankruptcy for different
household groups.41
For the average household, with $10,000 in home equity, the
probability of bankruptcy is identical in high and low exemption states. In either state,
the average probability of filing is 0.00016 percentage points. This explains why studies
using aggregate data were unable to find an empirical relationship between exemption
levels and bankruptcy. However, high asset households are more likely to file in high
41
All other factors are held constant at the mean of the data for the 6,996 households in the PSID. These
results are based on my baseline specification, seen in the second column of Table 3.
22
exemption states. Indeed a household with $100,000 in home equity is over 4 times more
likely to file in high exemption state. And a household with $200,000 in home equity is
27 times more likely to file in a high exemption state. Table 4 also confirms that low
income households are more likely to file in high garnishment states. In fact, households
with income of $10,000 are nearly 1.5 times more likely to file for bankruptcy in a high
garnishment state. Notice however, that for high income households, earning at least
$80,000, the predicted probability of bankruptcy does not depend on the garnishment
rate.
6.2 Policy Implications
Table 5 gives predicted changes in the probability of filing for bankruptcy that
result from hypothetical policy changes. Suppose first that every state enacts a high
exemption level, protecting at least $50,000 of home equity. The model predicts that the
average probability of bankruptcy would rise by 0.0019 percentage points.42 Since the
average probability of filing in my sample is 0.000423 percentage points, the model
predicts that the number of bankruptcy filings would increase by 44 percent per year.
Based on 822,590 bankruptcy filings per year in the United States (the figure for 2007),
this implies that approximately 363,000 additional bankruptcy filings would occur per
year.
Next, suppose every state enacts a high garnishment rate, allowing garnishment at
the federal limit of 25 percent. The model predicts that the average probability of
bankruptcy would rise by .00006 percentage points.43 Since the average probability of
filing in my sample is 0.000423 percentage points, the model predicts that the number of
bankruptcy filings would increase by 13 percent per year. Again, based on 822,590
bankruptcy filings per year in the United States (the figure for 2007), this implies that
approximately 111,000 additional bankruptcy filings would occur per year.
42
I calculate the change in each household’s probability of filing for bankruptcy. Note that the probability
of filing for bankruptcy does not change for households currently living in states with high exemption
levels.
43
Again, I calculate the change in each household’s probability of filing for bankruptcy. Note that the
probability of filing for bankruptcy does not change for households currently living in states with high
garnishment rates.
23
6.3 Levels of Debts and Assets
To date, no other bankruptcy paper has addressed whether debts or assets depend
on the state’s exemption and garnishment laws. Most earlier works have assumed that
these variables are exogenous; two notable exemptions (Repetto 1998; Pavan 2008)
examined whether assets and debts to depend on exemption laws. However, as detailed
in the model above, debts and assets depend on both exemption and garnishment laws.
The model predicts that households “borrow to save” in states with high
exemption levels. Recall that in states with a high exemption level, there is a greater
likelihood that a household will be able to retain its assets without having to repay its
debts. This creates an incentive for households to accumulate additional assets at the cost
of additional units of debts. Specifically, the model predicts that if A < E, dA/dE and
dD/dE will be zero and if A > E, dA/dE and dD/dE will be positive.
The model also predicts that households “borrow to save” in states with high
garnishment rates. The model predicts that if A < E, dA/dg and dD/dg will be positive
but if A > E, dA/dg and dD/dg will be zero.
Using data from the PSID, I examine whether assets and debts depend on state
exemption and garnishment laws.
Independent variables include dummy variables
denoting high exemption levels and high garnishment rates. Additionally, I interact these
dummy variables with a dummy variable N.E. which equals one if the household has nonexempt assets. I also include the household characteristics detailed above and regional
dummy variables.
Table 6 reports regression coefficient using seemingly unrelated
regression analysis (SUR). 44 Again, the errors are clustered at the state level.
Consistent with the model’s predictions, the coefficients on the dummy variable
denoting high exemption levels are statistically insignificant. The model also predicts
that the coefficients on the interaction term High Exemption * N.E. should be positive. It
is not surprising that the coefficient in column I does not have the expected sign; the
coefficient is likely biased downwards as the regression does not include information on
44
See Zellner (1962). Seemingly unrelated regression analysis allows the error terms to be correlated
across equations. As debts and assets are determined simultaneously, it is likely that the error term is
correlated across equations. Estimating a set of seemingly unrelated regressions jointly as a system will
yield more efficient estimates than estimating each of them separately.
24
the interest rate. Unfortunately, the PSID does not provide information on the interest
rate charged on unsecured debts.
The model also predicts that the coefficients on High Garnishment should be
positive while the coefficients on High Garnishment * N.E. should be negative. Again, it
is not surprising that the coefficients do not have the expected signs as the regressions do
not include information on the interest rate. This omitted variable causes the results in
Table 6 to be biased.
7. Robustness Checks
While my results are both reasonably signed and interesting, there are three
potential problems. First, my results may be driven by unobserved factors affecting
bankruptcy. To test whether this is the case, I re-run my regressions with state fixed
effects; I also perform a conditional logit regression. Another potential concern is that
the PSID sample is partially “contaminated” with bankrupt households; this concern is
tested in Section 7.3. Finally, in Section 7.4, I perform a multinomial logit regression.
7.1 State Fixed Effects
My baseline specification attests that state laws have a significant impact on a
household’s filing decision.
However, the reported coefficients could be biased by
spatially correlated variables that are not observed by the econometrician. For example,
households in a particular state could be influenced by the same social norms, legal
culture, and preferences to repay debt--- regional fixed effects may not control for these
unobserved characteristics.
To examine this issue, I re-run my regression with state fixed effects. It is
important to recognize that with the inclusion of state fixed effects I cannot estimate the
level effect of state bankruptcy laws. However, the interaction effects between the laws
and various household characteristics can still be identified.
As shown in Table 7, the inclusion of state fixed effects does not substantially
impact the magnitude of the other coefficients. While column I reports coefficients from
my baseline specification, column II reports coefficients when state fixed effects are
included. In addition, these state fixed effects have little explanatory value--- indeed
25
their partial r-squared is only 0.03. This important robustness check should alleviate any
concern that my results are driven by state-level omitted variables.45
7.2 McFadden Choice Model
Again, to alleviate concerns that my results are driven my unobserved
characteristics, in column III, I perform a conditional logit regression (also known as
McFadden’s choice model). Admittedly, while the coefficients are less significant, they
are consistent with earlier findings, high exemption levels encourage high asset
households to file for bankruptcy while high garnishment rates encourage low income
households to file for bankruptcy.
7.3 Contaminated Control Group
Because the 2007 PSID does not identify bankrupt households, I have, until now,
assumed that no household in the control sample filed for bankruptcy. However, because
1.2 percent of all households in the United States filed for bankruptcy in 2007, some
households in my control group have filed for bankruptcy.
Fortunately, Cosslett (1981) and Lancaster and Imbens (1996) develop an
estimation procedure to deal with the “contaminated control group” problem. Again, let
Bankrupt equal 1 if a household files for bankruptcy and X be a vector of explanatory
variables; X includes the legal, financial and demographic factors detailed in previous
tables. X is distributed according to f(X). The conditional probability of bankruptcy
given attributes X is denoted P=P(Bankrupt|X,β) and the marginal probability is given by
Q   P( Bankrupt | X ,  ) f ( X )dX .
Following Cosslett (1981) and Lancaster and Imbens (1996), if observations
1,…,N1 are from the bankruptcy sample and observations N1+1,…, N1+N2 are from the
PSID sample, then the log likelihood function is:
N1
N1  N 2
n 1
n 1
L(  , f )   ln P(in | X n ,  ) 

2
ln f ( X n )   N s ln{ dXf ( X )P( J ( s) | X ,  )}
s 1
45
Additionally, this robustness check should alleviate any concern that my results are driven by variations
in the costs of living across states.
26
where s is a sample indicator (s=1 denotes the choice-based sample of bankrupt
households and s=2 is a supplemental random sample containing unidentified households
that may be bankrupt) and J(s) is the set of chosen alternatives in subset s. When the
unknown distribution f(X) is approximated with weights w1 
w2 
N1
and
Q * ( N1  N 2 )
N2
, the resulting psuedolikelihood function is given by:
N1  N 2






N1  N 2


2
w1 P(i n | X n ,  )



Lˆ (  )   ln  2
   ln  ws P( J ( s) | X n ,  )  H 0 
n 1
  ws P( J ( s) | X n , )  H 0  n  N1 1  s 1

 s 1



N1
where H0=N1/N2. This formulation corrects for the problems of choice based sampling
and contaminated controls.
Results from this estimation procedure can be found in the fourth column of Table
7. As such a small fraction of controls are contaminated, it is unsurprising that the results
are robust to this alternative estimation procedure. Controlling for contamination does
not have a significant impact on the magnitude of the coefficients.
7.4 Multinomial Logit
The empirical analysis above employs a logit regression, estimating the
probability that a household files for bankruptcy (versus the alternative that they do not
file and choose either non-response or repayment in full). However, the theoretical
model supports a multinomial logit regression in which households are allowed to choose
between non-response, bankruptcy and full repayment. However, it was not possible to
estimate a multinomial logit regression with the dataset detailed above.
In this section, as a robustness check, I employ a multinomial logit regression
using the Child Development Supplement (CDS) of the PSID. In 2007, the CDS sample
consisted of 1,217 households with children aged 12 and younger. The supplement asked
households if, during the past year, they had fallen behind on their bills. Thus, with data
from this supplement, I am able to employ a multinomial logit regression, in which
households choose between non-response, bankruptcy, and repayment.
27
As seen in the fifth column of Table 7, although similar in magnitude, regression
coefficients are less significant. This may have occurred for several reasons. First, as
mentioned above, the sample size has decreased significantly, causing the standard errors
to rise. In addition, all the households in the CDS have children under the age of 12--- as
a result, there is less variation in the independent variables. This may also have caused
the standard errors to increase.
8. Conclusion
Previous work has provided little convincing evidence that state bankruptcy laws
impact the decision to file for bankruptcy. Using a newly collected household-level
dataset and econometric techniques to correct for choice-based sampling, I show that
state laws clearly affect who files for bankruptcy --- high property exemption levels will
encourage households with higher asset values to file for bankruptcy. In addition, I find
that high garnishment rates encourage households with low incomes to file for
bankruptcy. In particular, my theoretical model shows high garnishment rates encourage
low income households to choose bankruptcy over non-response; with its high costs,
bankruptcy is a more expensive form of default. A variety of robustness checks suggest
that my results are not driven by data problems such as omitted state-level variables.
To motivate my empirical analysis I develop a theoretical model in which
households choose between non-response, bankruptcy, and repayment of debt. Nonresponsive households are those that simply “walk away” from their debts. Although
previous models of bankruptcy have ignored the non-response option, it is relevant to the
problem studied in this paper because non-responsive households may be subject to wage
garnishment and asset seizure.
Although the primary purpose of my paper is to show that state laws influence
bankruptcy, my results raise important policy issues. While many Americans desire
some type of consumption insurance, 46 bankruptcy may not be the best option, depending
46
Bankruptcy provides consumption insurance by forgiving individual’s debts when their wealth or
earnings are low. However, it is not the only form of consumption insurance. Indeed, social safety nets,
such as food stamps, unemployment insurance, and workers compensation, all provide consumption
insurance through cash transfers. As detailed in Athreya and Simpson (2003), Athreya (2002), Fisher
28
on the household’s characteristics and legal environment. If state laws make it difficult
for poor households to file for bankruptcy, do they turn to other ways to manage income
shortfalls such as unemployment insurance or welfare? Further research should examine
the interaction between bankruptcy, income shocks, and other forms of consumption
insurance.
(2005), and Fisher (2001), both types of insurance programs come at nontrivial societal costs; bankruptcy
results in less credit and/or higher costs of credit while social safety nets are financed by income taxation.
29
A. Mathematical Appendix
This mathematical appendix details the theoretical model outlined in Section 4. Recall that the
household’s maximization problem is:
F/g
max U (Y1  A  D) 
A, D
 U ((1  g )Y2  min( A, E ))dY2 
max( A E , 0 )
 U (Y
2
0
D (1 r )  F
 U ( A  F )dY
2
 min( A, E )  F )dY2
F/g
1

max( A E , 0 )
 U (Y
2
 A  D(1  r ))dY2
D (1 r )  F
The household’s first order condition with respect to debt is:47
U (Y1  A  D)  (1  r  D
1
dr
)
U (Y2  A  D(1  r ))dY2  0
dD D (1r ) F
The household’s first order condition with respect to assets is:
F/g
 U (Y1  A  D) 

U ((1  g )Y2  A)dY2 *
0
d min( A, E )

dA
 [ D(1  r )  F  max( A  E ,0)]U ( A  F )  (1  D
max( A  E , 0 )

U ' (Y2  A  F )dY2 *
F/g
d min( A, E )
dA
1
dr
)
U (Y2  A  D(1  r ))dY2  0
dA D (1r )  F
As detailed in Section 4, the lender’s zero profit condition is given by:
F/g
 [ gY2  max( A  E,0)]dY2 
0
max( A E , 0 )
 max( A  E,0)dY2 
F/g
D (1 r )  F
 Y2 dY2 
max( A E , 0 )
1
 D(1  r )dY
2
 D(1  r f )
D (1 r )  F
The lender’s first order condition with respect to the interest rate is:
max( A  E,0) *
d max( A  E,0)
dD
 [ D  (1  r )
] * [1  D(1  r )]  0
dr
dr
A.1 Predictions about Assets, Debts and the Interest Rate
The first order conditions are totally differentiated and Cramer’s Rule is used to solve for
dA/dE, dD/dE, dr/dE, dA/dg, dD/dg, and dr/dg.
 ( x)
47
Leibnitz rule
 ( x)
d
d
d
dG
G( x, t )dt G( x,  ( x))
 G( x,  ( x))
 
dt is used to find the first order

dx  ( x )
dx
dx  ( x ) dx
conditions.
30
dA
[m12 m 23  m13 m 22 ]
dA d max( A, E )
dr

*
0
dE
dA
m11 (m 22 m33  m 23 m32 )  m12 (m 21m33  m 23 m31 )  m13 (m 21m32  m 22 m31 )
U ' ( A  F )[m12 m33  m13 m32 ] 
dA
U ' ( A  F )[m13 m31  m11m33 ] 
[m13 m 21  m11m 23 ]
dD d max( A, E )
dr

*
0
dE
dA
m11 (m 22 m33  m 23 m32 )  m12 (m 21m33  m 23 m31 )  m13 (m 21m32  m 22 m31 )
dA
[m11m22  m12 m21 ]
dr d max( A, E )
dr

*
0
dE
dA
m11 (m22 m33  m23m32 )  m12 (m21m33  m23m31 )  m13 (m21m32  m22 m31 )
U ' ( A  F )[m11m32  m12 m31 ] 
F/g
0 U " ((1  g )Y2  A)dY2 [m13m32  m12m33 ]
dA d min( A, E )

*
0
dg
dA
m11 (m22m33  m23m32 )  m12 (m21m33  m23m31 )  m13 (m21m32  m22m31 )
Y2
F/g
0 U " ((1  g )Y2  A)dY2 [m11m33 ]
dD d min( A, E )

*
0
dg
dA
m11 (m22 m33  m23m32 )  m12 (m21m33  m23m31 )  m13 (m21m32  m22 m31 )
Y2
F/g
Y2  U " ((1  g )Y2  A)dY2 [m11m32 ]
dr d min( A, E )
0

*
0
dg
dA
m11 (m22 m33  m23m32 )  m12 (m21m33  m23m31 )  m13 (m21m32  m22 m31 )
where
1
dr
m11  U " (Y1  A  D)  (1  r  D
)
U " (Y2  A  D(1  r ))dY2  0
dD D (1r )  F
m12  U " (Y1  A  D)  (1  r )(1  r  D
dr
)U ' ( A  F )
dD
1
 (1  r )(1  r  D
1
dr
dr
)  U " (Y2  A  D(1  r ))dY2 
U ' (Y2  A  D(1  r ))dY2  0
dD D (1 r )  F
dD D (1r )  F
1
dr
dr
m13  D(1  r  D
)U ' ( A  F )  D(1  r  D
)
U " (Y2  A  D(1  r ))dY2
dD
dD D (1r )  F
1

 U ' (Y
2
 A  D(1  r ))dY2  0
D (1 r )  F
31
F/g
m 21  U " (Y1  A  D) 

d min( A, E )

dA
U " ((1  g )Y2  A)dY2 *
0
[ D(1  r )  F  max( A  E ,0)]U " ( A  F )  U ' ( A  F ) *
max( A  E , 0 )

U " (Y2  A  F )dY2 *
F/g
d min( A, E )
dA
d max( A, E )
dA
1
 (1  D
dr
)
U " (Y2  A  D(1  r ))dY2  0
dA D (1r )  F
1
m 22
dr
dr
 U " (Y1  A  D)  D(1  r ) U ' ( A  F )  (1  r )(1  D )  U " (Y2  A  D(1  r ))dY2
dA
dA D (1 r )  F
1
dr

U ' (Y2  A  D(1  r ))dY2  0
dA D (1r )  F
1
m23  D 2
m31 
dr
dr
U ' ( A  F )  D(1  D )  U " (Y2  A  D(1  r ))dY2  0
dA
dA D (1 r )  F
dA d max( A, E )
*
0
dr
dA
m32  (1  r ) 2
dD
 2 D(1  r )  1  0
dr
and m33   D 2  2 D(1  r )
dD dD

0
dr dr
A.2 The Likelihood of Filing for Bankruptcy
As
detailed
in
Section
4,
households
file
for
bankruptcy
when
min[Y2  min( A, E)  F , A  F ]  Y2  A  Q(Y2 , A, D, r ) where
Q(Y2,A,D,r)=min[gY2+ max(A-E,0), D(1+r)].
Equivalently, households file for bankruptcy
when Y2  min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) .
From this condition I can make
several predictions about the likelihood of bankruptcy.
High exemption levels encourage high asset households to file for bankruptcy.
Let   min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) .
Then the probability that a

household files for bankruptcy is
 dY
2
. As shown above, dA/dE, dD/dE and dr/dE are all non-
0
32
negative. In addition, all three terms are increasing in assets. Thus, it is easy to see that dΩ/dE is
also non-negative and increasing in assets.
For instance, for low asset households (with A < E), dA/dE, dD/dE and dr/dE are all
equal to zero; as a result dΩ/dE is also equal to zero. However, for high asset households (with
A > E), dA/dE, dD/dE and dr/dE are all strictly positive; as a result dΩ/dE is also positive. In
other words, while high exemption levels do not impact low asset households, they encourage
high asset households to file for bankruptcy.
While high garnishment rates encourage low income households to file for bankruptcy.
Again, let   min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) so that the probability that

a household files for bankruptcy is
 dY
2
. Because dA/dg, dD/dg and dr/dg are non-negative and
0
decreasing
in
income,
it
can
be
shown
that
d d min[Y2  min( A, E ), A] dA d min[ gY2  max( A  E ,0), D(1  r )]
is non-negative and



dg
dg
dg
dg
decreasing in income.
The likelihood of bankruptcy is decreasing in A.
Again, if   min[Y2  min( A, E), A]  F  A  Q(Y2 , A, D, r ) , then it can be shown that
d
 0.
dA
The likelihood of bankruptcy is increasing in D.
Finally, it can be shown that
d
 0.
dD
33
References
Adler, B.E., B. Polak, and A. Schwartz. 2000. Regulating consumer bankruptcy: A
theoretical inquiry. Journal of Legal Studies 29 (2): 585-613.
Apilado, V.P., J.L. Dauten, and D.E. Smith. 1978. Personal bankruptcies. Journal of
Legal Studies 7 (2): 371-392.
Athreya, K.B. 2002. Welfare implications of the bankruptcy reform act of 1999. Journal
of Monetary Economics 49 (8): 1567-1595.
Athreya, K.B., and N.B. Simpson. 2003. Personal bankruptcy or public insurance?
Federal Reserve Bank of Richmond Working Paper #03-14, Federal Reserve
Bank of Richmond.
Buckley, F.H., and M.F. Brinig. 1998. The bankruptcy puzzle. Journal of Legal Studies
27 (1): 187-208.
Bucks, B.K., A.B. Kennickell, T.L. Mach, and K.B. Moore. 2009. Changes in U.S.
family finances from 2004 to 2007: Evidence from the Survey of Consumer
Finance. Federal Reserve Bulletin 95: A1-A56.
Cosslett, S.R. 1981. Maximum likelihood estimator for choice-based samples.
Econometrica 49 (5): 1289-1316.
Culhane, M.B., and M.M. White. 1999. Taking the new consumer bankruptcy model for
a test drive: means-testing real chapter 7 debtors. American Bankruptcy Institute
Law Review 7 (1): 28-75.
Dawsey, A.E., and L.M. Ausubel. 2004. Informal bankruptcy. University of Maryland
Working Paper, Department of Economics, University of Maryland at College
Park.
Domowitz, I., and T.L. Eovaldi. 1993. The impact of the bankruptcy reform act of 1978
on consumer bankruptcy. Journal of Law and Economics 36 (2): 803-835.
Domowitz, I., and R.L. Sartain. 1999. Determinants of the consumer bankruptcy
decision. Journal of Finance 54 (1): 403-420.
Fay, S., E. Hurst, and M.J. White. 2002. The household bankruptcy decision. American
Economic Review 92 (3): 706-718.
Fisher, J.D. 2001. The effect of transfer programs on personal bankruptcy. U.S.
Department of Labor Working Paper 346, U.S. Department of Labor.
34
——— 2005. The effect of unemployment benefits, welfare benefits, and other income
on personal bankruptcy. Contemporary Economic Policy 23 (4): 483-492.
Flynn, E. and G. Bermant. 2000. Pre-Bankruptcy planning limits means-testing impact.
American Bankruptcy Institute Journal 19 (2): 22-28.
Gropp, R., J.K. Scholz, and M.J. White. 1997. Personal bankruptcy and credit supply and
demand. Quarterly Journal of Economics 112 (1): 217-251.
Hynes, R.M, A. Malani, and E.A. Posner. 2003. The political economy of property
exemption laws. Journal of Law and Economics 47 (1): 19-43.
Lancaster, T., and G. Imbens. 1996. Case-control studies with contaminated controls.
Journal of Econometrics 71 (1): 145-160.
Lefgren, L., and F. McIntyre. 2009. Explaining the puzzle of cross-state differences in
bankruptcy rates. Journal of Law and Economics 52(2): 367-393.
Lefgren, L., F. McIntrye, and M. Miller. 2010. Chapter 7 or 13: Are client or lawyer
interests paramount? B.E. Journal of Economic Analysis and Policy 10(1).
Lehnert, A., and D.M. Maki. 2002. Consumption, debt, and portfolio choice: Testing the
effect of bankruptcy law. FEDS Working Paper Number 2002-14, Board of
Governors of the Federal Reserve.
Li, W., and P.-D. Sarte. 2006. U.S. consumer bankruptcy choice: The importance of
general equilibrium effects. Journal of Monetary Economics 53 (4): 613-631.
Livshits, I., J. MacGee, and M. Tertilt. 2007. Consumer bankruptcy: A fresh start.
American Economic Review 97 (1): 402-418.
Manski, C.F., and S.R. Lerman. 1977. The estimation of choice probabilities from choice
based samples. Econometrica 45 (8): 1977-1988.
Miller, L.E., and M.M. Miller. 2008. Repeat filers under the BAPCPA: A legal and
economic analysis. Norton Annual Survey of Bankruptcy Law and Practice. 2008:
509-534.
Nelson, J.P. 1999. Consumer bankruptcy and chapter choice: State panel evidence.
Contemporary Economic Policy 17 (4): 552-566.
Palank, J. 2008. Filing for bankruptcy becomes more costly: Paperwork swells, as do
lawyer fees, since 2005 overhaul. Wall Street Journal, July 31.
Pavan, M. 2008. Consumer durables and risky borrowing: The effects of bankruptcy
protection. Journal of Monetary Economics 55: 1441-1456.
35
Peterson, R.L., and K. Aoki. 1984. Bankruptcy filings before and after implementation of
the bankruptcy reform law. Journal of Economics and Business 36 (1): 95-105.
Repetto, A. 1998. Personal bankruptcies and individual wealth accumulation. mimeo,
CEA, Universidad de Chile.
Shiers, A., and D. Williamson. 1987. Nonbusiness bankruptcy and the law: Some
empirical results. Journal of Consumer Affairs 21 (2): 277-292.
Stanley, D.T., and M. Girth. 1971. Bankruptcy: Problem, process and reform.
Washington: Brookings Institute.
Sullivan, T.A., E. Warren, and J.L. Westbrook. 2000. The fragile middle class. New
Haven: Yale University Press.
Tabb, C.J. and J. McClelland. 2007. Living with the means test. University of Illinois
Law and Economics Research Paper # LE07-004.
Wang, H.-J., and M.J. White. 2000. An optimal personal bankruptcy system and
proposed reforms. Journal of Legal Studies 39 (2): 255-286.
Weiss, L., J. Bhandari, and R. Robins. 2001. An analysis of state-wide variation in
bankruptcy rates in the United States. Bankruptcy Developments Journal 17: 407424.
White, C.J. 2007. Making bankruptcy reform work: A progress report in year 2.
American Bankruptcy Institute Journal 26(5): 16-21.
White, M.J. 1987. Personal bankruptcy under the 1978 Bankruptcy Code: An economic
analysis. Indiana Law Journal 63 (1): 1-57.
——— 2005. Economic analysis of corporate and personal bankruptcy law. NBER
Working Paper #11536, National Bureau of Economic Research.
Zellner, A. 1962. An efficient method of estimating seemingly unrelated regressions, and
tests for aggregation bias. Journal of the American Statistical Association 57
(298): 348-368.
Zhu, N. Forthcoming. Household consumption and personal bankruptcy. Journal of Legal
Studies.
36
State
Table 1: State Exemption Laws and Garnishment Laws
Homestead
Garnishment Law*
Exemption
Wages that are Exempt
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
$5,000
$67,500
$150,000
$2,500
$50,000
$45,000
$75,000
$50,000
Unlimited
$10,000
$20,000
$50,000
$7,500
$15,000
Unlimited
Unlimited
$5,000
$25,000
$35,000
$0
$500,000
$3,500
$200,000
$75,000
$15,000
$100,000
$12,500
$350,000
$100,000
$0
$30,000
$50,000
$18,500
$80,000
$5,000
Unlimited
Federal Limit
$402.50
Federal Limit
Federal Limit
Federal Limit
Federal Limit
40 Times Minimum Wage
85 Percent
100 Percent
Federal Limit
80 Percent
Federal Limit
85 Percent or 45 Times Minimum Wage
Federal Limit
40 Times Minimum Wage
Federal Limit
Federal Limit
Federal Limit
40 Times Minimum Wage
Federal Limit
Federal Limit
40 Times Minimum Wage
40 Times Minimum Wage
Federal Limit
90 Percent
Federal Limit
85 Percent
Federal Limit
50 Times Minimum Wage
90 Percent
40 Times Minimum Wage
90 Percent
Federal Limit
40 Times Minimum Wage
Federal Limit
Federal Limit
37
Table 1 Continued: State Exemption Laws and Garnishment Laws
State
Homestead
Garnishment Law*
Exemption
Wages that are Exempt
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming
$25,000
$0
$200,000
$50,000
$30,000
$5,000
Unlimited
$20,000
$75,000
$5,000
$40,000
$25,000
$40,000
$10,000
40 Times Minimum Wage
100 Percent
Federal Limit
100 Percent
80 Percent
Federal Limit
100 Percent
Federal Limit
75 Percent Above Minimum Wage
Federal Limit
40 Times Minimum Wage
80 Percent
80 Percent
Federal Limit
* "Federal Limit" indicates that 75 percent or 30 times the federal minimum wage per week
is exempt from garnishment.
38
Table 2: Descriptive Statistics
Bankruptcy Sample
Income in Year 1
Income in Year 2
Change in Income
Home Equity
Unsecured Debt
Known Demographics
Family Size
Single
Married
Separated
Divorced
Widowed
Imputed Demographics
25-34 Years Old
35-44 Years Old
45-54 Years Old
55-64 Years Old
65-74 Years Old
PSID Sample
$36,976.54
(998.58)
$31,886.64
(714.12)
-$5,089.90
(811.38)
$9,833.25
(1,696.84)
$50,876.63
(2,028.49)
$56,574.44
(1,126.27)
$69,064.64
(2,397.39)
$12,490.20
(2,123.69)
$95,416.07
(2,858.05)
$8,017.33
(321.61)
2.51
(0.04)
0.32
(0.01)
0.43
(0.01)
0.06
(0.01)
0.14
(0.01)
0.03
(0.00)
2.35
(0.02)
0.22
(0.01)
0.48
(0.01)
0.03
(0.00)
0.17
(0.01)
0.09
(0.01)
0.25
(0.01)
0.46
(0.01)
0.17
(0.01)
0.03
(0.00)
0.02
(0.00)
0.18
(0.01)
0.20
(0.01)
0.21
(0.01)
0.16
(0.01)
0.09
(0.01)
39
Table 2 Continued: Descriptive Statistics
Bankruptcy Sample
PSID Sample
Age 75 Years and Above
High School Degree
Some College
Associates Degree
Bachelor's Degree
Master's Degree
Professional/Doctorate Degree
Sample Size
0.04
(0.00)
0.40
(0.00)
0.18
(0.00)
0.08
(0.00)
0.16
(0.00)
0.06
(0.00)
0.02
(0.00)
0.09
(0.01)
0.34
(0.01)
0.19
(0.01)
0.04
(0.00)
0.18
(0.01)
0.04
(0.00)
0.02
(0.00)
1,694
6,996
Standard errors are in parentheses.
40
Table 3: Probability of Bankruptcy
(I)
High Exemption
Income Squared
Change in Income
Unsecured Debt
Home Equity
(IV)
(V)
(VI)
-0.042
-0.198
-0.18
-0.085
(0.224)
(0.210)
(0.228)
[-1.25E-8]
[-4.90E-8]
[-7.05E-8]
[-2.10E-8]
0.175***
0.160***
0.179***
(0.061)
(0.056)
(0.065)
[4.57E-8]
[6.52E-8]
[4.52E-8]
0.254
0.442*
0.428*
0.515**
0.424
(0.242)
(0.239)
(0.249)
(0.244)
(0.268)
[7.76E-8]
[1.18E-7]
[1.78E-7]
[1.63E-7]
[1.01E-7]
-0.057
-0.054
-0.057
-0.036
(0.043)
(0.042)
(0.048)
(0.042)
[-1.48E-8]
[-2.19E-8]
[-1.75E-8]
[-8.58E-9]
-0.321***
-0.298***
-0.219***
-0.325***
-0.343***
-0.283***
(0.032)
(0.032)
(0.031)
(0.035)
(0.034)
(0.054)
[-9.72E-8]
[-7.77E-8]
[-8.96E-8]
[-1.00E-7]
[-8.25E-9]
[-7.14E-8]
0.001***
0.001***
0.000***
0.001***
0.003***
0.000***
High Garnishment Rate *
Income
Income
(III)
(0.225)
High Exemption *
Home Equity
High Garnishment Rate
(II)
0.000
0.000
0.000
0.000
0.000
0.000
[1.72E-10]
[1.37E-10]
[1.61E-10]
[1.75E-10]
[6.15E-10]
[1.26E-10]
-0.017***
-0.018***
-0.018***
-0.018***
-0.018***
-0.017***
(0.004)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
[-5.06E-9]
[-4.64E-9]
[-7.23E-9]
[-5.42E-9]
[-4.28E-9]
[-4.34E-9]
0.154***
0.168***
0.165***
0.167***
0.161***
0.168***
(0.014)
(0.016)
(0.015)
(0.016)
(0.015)
(0.017)
[4.67E-8]
-0.192***
[4.39E-8]
-0.248***
[6.76E-8]
-0.256***
[5.17E-8]
[3.87E-8]
-0.246***
[4.22E-8]
-0.252***
(0.043)
(0.058)
(0.052)
(0.054)
(0.063)
[-5.82E-8]
[-6.48E-8]
[-1.05E-7]
[-5.92E-8]
[-6.35E-8]
41
Table 3 Continued: Probability of Bankruptcy
(I)
(II)
(III)
(IV)
Non-Exempt Home Equity
(V)
(VI)
-0.228***
(0.063)
[-7.05E-8]
Homestead Exemption Level
0.085
(0.000)
[-2.04E-9]
Homestead Exemption Level * Equity
0.000***
(0.000)
[-2.03E-9]
Some Garnishment is Legal
0.676**
(0.270)
[1.39E-7]
Some Garnishment is Legal * Income
-0.046
(0.055)
[-1.16E-8]
Demographics
Fixed Effects
Known &
Unknown
Region
Known &
Unknown
Region
Known
Region
Known &
Unknown
Region
Known &
Unknown
Region
Known &
Unknown
Region
Robust standard errors are in parentheses, clustered at the state level.
Marginal effects are in brackets.
As detailed in Table 2, known demographic variables include family size and marital status while age and
education are imputed demographic variables. Income, home equity and unsecured debt are measured in
$10,000.
*** significant at 1 percent; ** significant at 5 percent; * significant at 10 percent
42
Table 4: Predicted Probabilities (in Percentage Points)
by Legal and Demographic Factors
Low
High
Points Along Home Equity Distribution
Exemption Exemption
State
State
10th Percentile: $0 in Home Equity
25th Percentile: $0 in Home Equity
50th Percentile: Approximately $10,000 in Home Equity
75th Percentile: Approximately $100,000 in Home Equity
90th Percentile: Approximately $200,000 in Home Equity
0.00021
0.00021
0.00016
0.00002
0.00000
0.00017
0.00017
0.00016
0.00008
0.00004
Low
High
Garnishment Garnishment
State
State
Points Along the Income Distribution
10th Percentile: Approximately $10,000 in Income
25th Percentile: Approximately $25,000 in Income
50th Percentile: Approximately $50,000 in Income
75th Percentile: Approximately $80,000 in Income
90th Percentile: Approximately $125,000 in Income
0.00013
0.00008
0.00004
0.00002
0.00000
0.00019
0.00011
0.00005
0.00002
0.00000
This table contains the predicted probabilities of bankruptcy differentiated by legal and
demographic factors. All other household characteristics are held fixed at their mean value.
43
Hypothesized Change
Table 5: Policy Implications
Percentage Point Percentage Change
Additional
Marginal Effect
in Filing Rate
Filings Per Year
All states impose high exemption levels
(at least $50,000 of home equity is protected)
All states impose high garnishment rates
(25% of wages can be garnished)
0.00019
44.2
363,651
0.00006
13.5
110,845
This table contains the predicted probabilities of bankruptcy under hypothetical policy changes, holding all other household
characteristics at their mean value.
44
Table 6: Level of Debts and Assets
(I)
(II)
Debts
Assets
High Exemption
-0.048
(0.052)
-0.416**
(0.161)
-0.098*
(0.055)
0.086
(0.078)
0.014*
(0.008)
0.000***
(0.000)
Known &
Imputed
SUR
Full
Region
High Exemption * N.E.
High Garnishment
High Garnishment * N.E.
Income
Income Squared
Demographics
Regression
Sample
Fixed Effects
1.147
(1.426)
13.768***
(4.769)
-3.453**
(1.610)
10.280**
(1.013)
0.758***
(0.142)
-0.002*
(0.001)
Known &
Imputed
SUR
Full
Region
Robust standard errors are in parentheses, clustered at the state level.
As detailed in Table 2, known demographic variables include family size
and marital status while age and education are imputed demographic
variables. Income, home equity and unsecured debt are measured in
$10,000.
*** significant at 1 percent; ** significant at 5 percent; * significant at
10 percent
45
Table 7: Robustness Checks
(I)
(II)
(III)
(IV)
(V)
Baseline
State Fixed McFadden Contaminated Multinomial
Specification
Effects
Choice
Controls
Logit
High Exemption
High Exemption * Home Equity
High Garnishment
High Garnishment * Income
Income
Income Squared
Change in Income
Unsecured Debt
Home Equity
Demographics
Regression
Fixed Effects
-0.198
(0.224)
0.175***
(0.061)
0.442*
(0.239)
-0.057
(0.043)
-0.298***
(0.032)
0.001***
(0.000)
-0.018***
(0.003)
0.168***
(0.016)
-0.248***
(0.058)
Known &
Imputed
-0.062
(0.045)
-0.364***
(0.052)
0.001***
(0.000)
-0.021***
(0.004)
0.243***
(0.042)
-0.161**
(0.075)
Known &
Imputed
Logit
Logit
Region
State
0.175***
(0.077)
-0.197*
(0.116)
0.040**
(0.017)
0.244
(0.160)
-0.003
(0.036)
-0.329***
(0.033)
0.001***
0.000
-0.018***
(0.003)
Known &
Imputed
McFadden
Choice
Region
-0.198
(0.124)
0.175***
(0.042)
0.441**
(0.178)
-0.056
(0.039)
-0.298***
(0.033)
0.001***
(0.000)
-0.018***
(0.003)
0.168***
(0.013)
-0.248***
(0.038)
Known &
Imputed
Logit
Region
-0.433
(0.469)
0.070
(0.062)
0.462
(0.302)
0.075
(0.061)
-0.325***
(0.078)
0.001
(0.001)
-0.175***
(0.056)
0.281***
(0.064)
-0.218***
(0.052)
Known &
Imputed
Multinomial
Logit
Region
Robust standard errors are in parentheses, clustered at the state level.
As detailed in Table 2, known demographic variables include family size and marital status while age and education are imputed
demographic variables. Income, home equity and unsecured debt are measured in $10,000.
In column IV, the conditional logit regression (also known as McFadden's choice model) is computed using STATA's clogit
command. This analysis differs from the regular logit regression in that the data are stratified and the likelihoods are computed
relative to each stratum.
*** significant at 1 percent; ** significant at 5 percent; * significant at 10 percent
46
Figure 1: Second Period Consumption and Filing Decision
C2
A-F
(1- g)F+min(A,E)
g
min(A,E)
F
g
min(A-E,0)
D(1  r )  F
File for Bankruptcy
Repay in Full
Non-response
47
Y2