Do Chapter 11 Bankruptcy Filings Improve Firms’ Operating
Performance?
Varouj Aivazian∗ Simiao Zhou†
This draft: December 15, 2006
Abstract
This paper investigates the impact of Chapter 11 bankruptcy filings on the operating per-
formance of public U.S. firms. The paper generates from a comparison group of non-filing
firms a control group that has pre-filing characteristics similar to those of bankruptcy filing
firms. Comparison is made only between filing firms and matched non-filing firms with similar
pre-filing characteristics. This separates the pure effect of the bankruptcy filing from the effect
of other economic factors since presumably other economic factors affect similar firms alike. To
implement the matching method, this paper identifies ten factors that determine firms’ filing de-
cisions, such as profitability, liquidity, equity value, level and type of debt, as well as measures
reflecting agency and bargaining problems. Both traditional matching and propensity-score
matching methods are implemented. The different matching estimates obtained turn up to
be quite consistent and show that for our sample of US public firms, on average, Chapter 11
bankruptcy filings had net benefits and improved firms’ operating performance as measured by
sales and cash-flows normalized by total assets.
∗Joseph L. Rotman School of Management and Department of Economics, University of Toronto; Email:[email protected]
†Department of Economics, University of Toronto; Email: [email protected]
1
This paper investigates empirically the impact of Chapter 11 bankruptcy filings on the operating
performance of a sample of U.S. public firms. In the conventional view of the financial economics
literature, significant costs are associated with the act of bankruptcy filing. Those costs are termed
“direct” and “indirect”, with legal and administrative costs categorized under the former and lost
sales and foregone profits under the latter. An ideal bankruptcy law imposes a certain structure
on the negotiations to reduce bargaining costs; it forces firms to use assets more efficiently and to
terminate unprofitable projects. Formal bankruptcies are usually accompanied by comprehensive
organizational changes in management and governance structure, changes which can potentially
create value by improving the allocation of resources within the firm. An alternative to formal
bankruptcy filing is private negotiations with creditors to attenuate the costs associated with the
formal process. But, private negotiations may entail bargaining problems and transaction costs
that the formal process ostensibly reduces.
Ultimately, an efficient bankruptcy system induces an optimal rate of exit and entry of firms
by leading to the reorganization (liquidation) of companies whose going concern value exceeds (is
less than) their liquidation value. Whether existing formal bankruptcy systems meet this efficiency
benchmark is difficult to test; instead, the focus of this paper is the narrower one of testing the
impact of reorganization under Chapter 11 filing on firm performance. If firms perform better
(worse) after formally reorganizing than they would by not filing for bankruptcy, then bankruptcy
filing has net benefits (costs). Thus, we address a narrower question about the efficiency of the
formal bankruptcy process, whether it induces superior performance by formally reorganized firms
compared to otherwise similar (or matched) firms that do not formally file for bankruptcy.
Extant empirical work estimating the impact of bankruptcy filings on firm performance com-
pares the values of certain operating measures before and after bankruptcy filing with adjustments
for industry median performance. Intuitive as it is, this procedure suffers from some problems.
It does not adequately control for pre-filing individual firm characteristics, and thus does not dis-
criminate whether the estimated effects are due to the bankruptcy filing itself or due to poor firm
performance characteristics and prospects. Controlling only for industry median confounds the ef-
fect of the bankruptcy filings itself with those of other industry or firm-specific economic factors. To
address these problems, one needs an appropriate control group of firms that have similar pre-filing
characteristics to those of the bankrupt firms and that can best predict the counterfactual, namely,
2
what would have happened to the bankrupt firms had they not filed for bankruptcy? To find such a
control group, this paper utilizes matching methods to estimate the impact of bankruptcy filings.
By construction, the matching method explicitly controls for pre-filing firm attributes. It gener-
ates from a comparison group of non-filing firms a control group that has pre-filing characteristics
similar to those of bankruptcy filing firms. Thus, the comparison is between filing firms and
matched non-filing firms with similar pre-filing characteristics. This separates the pure effect of the
bankruptcy filing from the effect of other economic factors since presumably other economic factors
affect similar firms alike. To implement the matching method, this paper identifies ten factors
that determine firms’ bankruptcy filing decisions, such as profitability, liquidity, equity value, level
and type of debt, as well as measures reflecting agency and bargaining problems. The traditional
as well as propensity-score matching methods are implemented. The different matching estimates
that we obtain are consistent and show that for our sample of US public firms, on average, Chapter
11 bankruptcy filings had net benefits and improved firms’ operating performance. Our results
are robust with respect to different filing periods and also when matching is conditional on firm
attributes at different pre-filing years.
Our results do not show that formal bankruptcy filing is always a better mechanism than
alternatives such as informal reorganization. Instead, the question of interest is whether firms
that file perform better after bankruptcy than they would have without filing, and our results
suggest that the formal bankruptcy system does help these firms. It may be that certain firm
characteristics systematically affect the decision to file formally instead of reorganizing informally,
e.g., the decision may depend on the number and homogeneity of creditors, so that potentially
bankrupt firms may self-select into formally versus informally reorganized categories. To deal with
the self-selection issue, the matching method controls for these firm characteristics and assumes that
the unobservables will not affect filing decisions systematically, so that the control group consists
of firms that are similar to filing firms in terms of their systematic characteristics but do not file
for whatever reasons. The control group may or may not include informally reorganized firms. Our
data do not allow us to determine whether firms in the control group were informally reorganized.
The rest of the paper is organized as follows. Section I provides a brief discussion of financial
distress and the bankruptcy law, while Section II reviews the theories of bankruptcy and reorgani-
zation deducing testable implications for evaluating firms’ bankruptcy filing decisions. Section III
3
discusses weaknesses in existing empirical estimations of the impact of bankruptcy, pointing out
that these weaknesses can be overcome at least partially by matching methods, which are intro-
duced in Section IV. Section V identifies determinants of the bankruptcy filing decisions used in our
matching method, Section VI describes the dataset and provides empirical results, and Section VII
concludes the paper.
I. Financial Distress and Bankruptcy Law
What is financial distress for a corporation? We define it as the inability of the firm to meet its
contractual obligations. The firm can in general resolve financial distress in two ways, informally or
formally. In the former case, the firm makes private arrangements to renegotiate with its creditors
and to restructure the debt, or it chooses to shut down and to liquidate its assets. In the latter case,
the firm has the same two options, but the process itself is governed by the legal and regulatory
rules embodied in the Bankruptcy Code. Since this paper focuses on the estimation of the impact
of bankruptcy filing on firms in the US, some of the key features of the US bankruptcy code merit
brief discussion.
The liquidation and restructuring processes are governed by Chapters 7 and 11 of the US
Bankruptcy Code, respectively. When a firm files for Chapter 7, the bankruptcy court appoints
a trustee who shuts the firm down, sells its assets and turns the proceeds over to the court for
payment to creditors. The distribution of payments is made under the so called “Absolute Priority
Rule” (APR). The APR specifies that claims are paid in full in the following order: (1) secured
creditors’ claims; (2) administrative expenses of the bankruptcy process itself, including court costs,
lawyers’ fees, trustee expenses; (3) claims taking statutory priority, including tax claims, rent claims,
consumer deposits, and unpaid wages and benefits that accrued before the bankruptcy filing; (4),
unsecured creditors’ claims, including those of trade creditors and long-term bondholders; (5) claims
of equity holders.1
Instead of filing for liquidation, the firm can instead file for reorganization under Chapter 11,
which provides an opportunity for the firm to get relief from its creditors while it attempts to
restructure its debt and improve its financial situation. The Chapter 11 filing triggers an “automatic
stay,” which essentially freezes all of the firm’s debts and obligations to creditors and all attempts1see White (1989)
4
to collect on those obligations until a reorganization plan is formulated and ultimately confirmed
by the court.
In Chapter 11, the existing managers of the firm usually remain in control and the firm continues
to operate as a “debtor-in-possession.” The firm is exclusively entitled to propose a reorganization
plan during the first 120 days following the filing of a bankruptcy petition and has an additional 60
days to obtain acceptance by the creditors. After the initial 180 days have elapsed, if acceptance
has not been obtained, other parties can file a reorganization plan.
The Bankruptcy Code requires that each party under a reorganization plan receive at least as
much as it would in a Chapter 7 liquidation that is governed by APR. So, in principle, each creditor
class is compensated with the face value of its pre-bankruptcy claims only after all senior creditor
classes are paid in full. For a plan to be accepted, it is required that all classes of creditors and
equity as a class vote to approve the plan. For each class of creditors, section 1126(c) of the Code
requires a voting margin in favor of the plan to be at least two-thirds in amount of claims and
one-half in number of claimants. For equity, section 1126(d) requires a relative unanimity in voting
of at least two thirds.
To break a deadlock, the court can unilaterally impose, or “cram down” the plan on dissenting
classes if the plan is considered to be “fair and equitable”. Application of the fair and equitable
standard is taken by the court to mean that the plan satisfies the Absolute Priority Rule to the
asset value determined by the court.
If a plan of reorganization is confirmed by the court, and implemented, then the firm emerges
from bankruptcy protection. If, on the other hand, no progress is made toward completion of the
Chapter 11 reorganization, then normally some creditors petition the bankruptcy judge to order a
shift of the firm’s bankruptcy filing to a Chapter 7 liquidation.
II. Theories of Bankruptcy-reorganization and Testable Implications
When a distressed firm files for bankruptcy protection, this involves legal and professional ex-
penses such as administrative fees and fees for lawyers, trustees and accountants. Given bankruptcy
filing fees, why would a firm not negotiate with creditors privately to avoid such costs? Presumably
because private negotiations also entail costs.
5
Haugen and Senbet (1978) offer an arbitrage argument to show that firms will not resort to a
bankruptcy filing if the market is functioning well . They reason that in a well functioning market,
any potential bankruptcy will be capitalized into the firm’s market value and will reduce that value,
yielding an arbitrage opportunity. In such a world, equity holders, debt holders or outside investors
have incentives to purchase the firm as a whole at the prevailing market price. Hence, the firm will
not resort to a bankruptcy filing. The only costs of bankruptcy would be the small transaction
costs in the market. This reenforces the question of why firms file for a costly formal bankruptcy
if private arrangements have fewer costs. One explanation is that private arrangements may suffer
from severe bargaining problems as different claimholders have conflicting interests.
Recontracting among claimholders can resolve bankruptcy as analyzed by Aivazian and Callen
(1980, 1983). They argue that recontracting among shareholders and debtholders can potentially
overcome any inefficiency generated by bankruptcy, as long as bargaining costs are negligible. They
model equity and debtholders as players in a cooperative game, bargaining over any surplus resulting
from the difference between the firm’s going-concern value and its liquidation value. Using the
theory of the core, they also analyze the recontracting game and argue that the core of the game may
be empty in some circumstances. They claim that contractual and institutional mechanisms, such
as the formal bankruptcy code, may reduce transaction costs and overcome empty-core problems,
thus helping to generate efficient recontracting outcomes in bankruptcy.
Brown (1989) considers the holdout problem in a private-negotiation game, where there is a set
of mutually advantageous reorganization plans and another plan that makes at least one class of
claimants better off. Since any class of claimholders can veto the outcome, unanimous agreement
on a reorganization plan is difficult to achieve. The structure imposed by the Bankruptcy Code
mitigates the holdout problem yielding a unique solution to the reorganization process.
Giammarino (1989) considers the problems generated for the firm’s reorganization choice by
asymmetric information. In particular, the equityholders and the court are assumed to be aware of
the firm’s type, while the debtholders only have prior beliefs about the distribution of firm types.
If private renegotiations fail, the informed court can use its discretion to impose a reorganization
plan. Giammarino shows that in some equilibria the debtholders reject the equityholders’ plan and
proceed with a costly bankruptcy filing for resolving financial distress.
What the above papers highlight is that information and bargaining problems could frustrate
6
and make costly private reorganization plans. These costs include, in addition to transaction costs
of bargaining, those due to the disruption of the firm’s normal operations and the ensuing adverse
impact on the firm’s performance. Resort to a formal bankruptcy filing might mitigate these
problems and reduce the associated costs.
To see the potential benefits of a formal bankruptcy process note that in a private workout, in
order for a reorganization plan to be accepted it must get unanimous approval from all impaired
classes of creditors. Furthermore, most indentures include a cross-default provision that stipulates
that an issue is in default whenever any other issue is in default, and thus, approval of a plan
generally requires the approval of all the claimholders in all classes. The potential holdout problem
is, therefore, quite severe in a private workout because of the veto power of individual creditors.
A Chapter 11 filing can mitigate this holdout problem because it has restrictive voting rules that
make it easier to get a reorganization plan accepted. Different classes of claims can be consolidated.
Provision §1122 of the Bankruptcy Code allows claims to be placed in the same class as long as they
are substantially similar, and acceptance of the plan requires an affirmative vote by only a simple
majority of the claimholders. Provision §1125 of the Bankruptcy Code under Chapter 11 also helps
attenuate asymmetric information problems by requiring disclosure of adequate information before
soliciting acceptance of a reorganization plan. A written disclosure must be approved by the court,
and transmitted to claimholders together with the plan.
Chapter 11 rules also help improve the firm’s operating performances in other ways. First, the
automatic stay provision protects the firm from creditor interference during reorganization, thus
reducing disruptions of its business. Second, the Bankruptcy Code allows the firm to issue new
debt that is senior to debt incurred prior to filing. Such debtor-in-possession financing is valuable
for firms that face stringent liquidity constraints and cannot easily get outside financing in a private
workout. Yost (2002) mentions two additional benefits of Chapter 11 filing. If a firm restructures
out-of-court and exchanges a higher face value debt for lower face value debt, the difference between
the face values is considered taxable income. Similarly, if a firm restructures out-of-court and offers
equity to debtholders and if ownership by existing shareholders becomes less than 50% of original
ownership, the firm forfeits its accumulated net operating loss carryforwards. But the tax liability
on cancellation of indebtedness income and the forfeiture of net operating loss carryforwards do not
apply to a Chapter 11 filing.
7
Finally, bankruptcy filing offers an opportunity for the firm to restructure completely and to get
a fresh start. At the same time, it signals to the market information about the firm’s prospects and
enables a re-evaluation of the firm. As argued in Wruck (1990), some firms in bankruptcy filing un-
dergo dramatic organizational changes as part of their recovery and refocus their strategy. Without
outside intervention, management may fail to change to an optimal strategy; hence, bankruptcy
filing can force managers to undertake value-enhancing organizational changes that they would not
otherwise have undertaken.
The foregoing discussion suggests that bankruptcy law mitigates bargaining and asymmetric
information problems in the renegotiation process, and can force managers to make value-increasing
operational changes. It is therefore quite reasonable to argue that firms are actually better off under
Chapter 11 than they would be otherwise. This leads to the paper’s central hypothesis: Chapter 11
bankruptcy filing has no significant costs or adverse affects on firms’ operating performance; that
firms, on average, perform at least as well with Chapter 11 filing as they would without such filing.
The paper empirically evaluates the impact of Chapter 11 filing on firm performance. While
extant empirical results of the impact Chapter 11 filing have been interpreted as estimates of indi-
rect bankruptcy costs, we do not find it fruitful to distinguish empirically between the direct and
indirect costs (or benefits) of bankruptcy. We think that the empirical evaluation should simply
reflect the overall effect of bankruptcy filing on firm performance. Filing fees reflect commitment
by firms to legal protection and reliance on the bankruptcy process. This process entails incurring
administrative fees and professional expenses, changing the firm’s operating and management strat-
egy, and restructuring governance. A simple and better way to quantify the impact of bankruptcy
is to sidestep the process itself and to concentrate on outcomes, by investigating how firms fare
at different points after bankruptcy. The firm’s post-bankruptcy performance should reflect all
the costs and benefits, including filing expenses, related to the bankruptcy process. To evaluate
the overall impact of bankruptcy, it is necessary to separate the pure bankruptcy filing effect from
effects inherent in pre-filing firm performance characteristics and other economic indicators. This,
in turn, entails adequately controlling for firms’ (pre-filing) characteristics, a factor not properly
taken into account in the extant literature.
8
III. Extant Empirical Estimations of Bankruptcy Costs
Altman (1984) estimates indirect bankruptcy costs using measures based on the concept of fore-
gone profits. Expected profits are first estimated by a regression technique using average industry
sales figures as the benchmark, and then expected profits are compared with actual profits. The
differences are considered estimates of indirect bankruptcy costs. This process is vulnerable to two
problems. First, the estimated profits figure represents the “would-be” outcome for the bankrupt
firm had it not filed for bankruptcy; no underlying theory predicts any particular functional form for
this outcome variable. It may be preferable to estimate the “would-be” outcome non-parametrically.
Second, the regression does not isolate the pure impact of bankruptcy filing. For example, suppose
there is an industry downturn due to an economic factor common for all firms in a forecast year;
the distressed firm’s actual profits will then reflect the impact of the bankruptcy filing itself and
the economic downturn as well. Thus, a simple comparison of predicted profits with actual profits
tends to confound those two kinds of impacts.
Andrade and Kaplan (1998) study 31 companies that made highly leveraged transactions and
subsequently became financially distressed. To estimate the costs of financial distress, they con-
structed firm-performance measures and computed the changes in each measure from the onset of
financial distress to the resolution of the distress. One set of measures consisted of operating perfor-
mance indicators, normalized by sales or assets. They also adjusted their measures by industry or
market performance indicators and found that the indirect costs of financial distress, as captured
by changes in operating performance measures, ranged from 7 to 17 percent, with a decline in
operating and net cash flow margins from the year before distress to the year after the of resolution
of distress.
Maksimovic and Phillips (1998) used plant-level data to investigate indirect bankruptcy costs for
manufacturing firms that filed for reorganization under Chapter 11. One of their objectives was to
estimate these costs by measuring changes in the plant-level Total Factor Productivity (TFP) and
in operating cash flows between pre-Chapter 11 and post-Chapter 11 periods. They adjusted their
measures by the industry median and found that a filing firm’s performance, on average, did not
differ significantly from its industry counterparts in low-demand industries, while it deteriorated in
high-demand industries.
A common strategy employed in the latter two papers was to compare firm performance measures
9
before and after the bankruptcy filing; the only control for comparison was the industry median.
There are two concerns with this empirical strategy. First, it does not adequately control for pre-
filing firm characteristics. Hence it cannot determine whether the estimated effects are due to the
bankruptcy filing itself or due to poor firm performance. Thus, the differenced operating outcomes
before and after bankruptcy may incorporate the influence of a deteriorating operating performance
beforehand as well as the effect of the bankruptcy filing. Adjusting only for the industry median does
not resolve this problem since the industry median may have quite different pre-filing characteristics
from those of bankruptcy filing firms. This problem can be mitigated in some specific samples such
as the one employed by Andrade and Kaplan (1998). They are able to identify high leverage as
the primary reason for financial distress while other attributes such as operating performance play
a negligible role. In general, however, we need other methods of addressing this problem. Second,
controlling only for the industry median confounds the effects of the bankruptcy filing itself with
those of other economic factors during the course of a bankruptcy filing. The simple unidimensional
control will not work unless the common economic factors uniformly affect all firms in the industry.
In reality, however, the same economic factor may have different effects on different firms because
of distinct firm-specific characteristics.
To address the above problems, one needs to find an appropriate control group for comparison.
Ideally, the control units must mimic the counterfactual, namely, what would have happened had
the bankrupt firm not filed for bankruptcy? Comparing the “actual” outcome with the “would-be”
(counterfactual) outcome eliminates the influence of all other economic factors, isolating the pure
effect of a bankruptcy filing. The matching method is a way of finding such control units, as it
eliminates or at least reduces, the estimation bias resulting from an inadequate control for firm
attributes.
IV. Evaluating the Impact of Bankruptcy Filing with Matching Methods
The empirical work carried out in this paper evaluates the impact of Chapter 11 filing on firms’
operating performance. This task is accomplished by the employment of matching methods, tech-
niques commonly used in the social program evaluation literature. In the context of a bankruptcy
filing, the firm’s filing decision is labeled as a treatment. A filing firm i is called a treatment unit
in the treatment group and labelled as Di = 1. A non-filing firm is called a comparison unit in
10
the comparison group with Di = 0. Each firm has a vector of pre-filing characteristics, referred
to as covariates, and denoted as Xi. The operating outcomes for filing and non-filing firms are
denoted by Yi1 and Yi0, respectively. Given these notations, the impact of a Chapter 11 filing on a
firm’s performance is given by (Yi1−Yi0|Di = 1). The second part of the expression, Yi0|Di = 1, is
the counterfactual outcome of a filing firm, capturing what would have happened had it not filed
for bankruptcy. As argued in Section II, the central question addressed in this paper is whether
firms perform, on average, at least as well as they would have had they not filed for bankruptcy.
Hence, the parameter of interest is given by E[Y1 − Y0|D = 1],which is commonly called the Av-
erage Treatment effect on the Treated (ATT); it captures the average difference between the effect
of the treatment on the treated units and the effect if they had not received the treatment. The
mean, E[Y1|X, D = 1], can be identified from data on filing firms. The mean E[Y0|X, D = 1] is
called the counterfactual mean, and cannot be identified from data. It seems intuitive to replace
E[Y0|X, D = 1] with E[Y0|X, D = 0] for which data are available. However, a naive comparison
between E[Y1|X,D = 1] and E[Y0|X,D = 0] leads to a selection bias.
Matching is a way of estimating the treatment effect parameter using observational data. It
reduces the selection bias by constructing a control group of “comparable” units. The rest of this sec-
tion outlines the assumptions underlying matching and its empirical implementation. Appendix A
contains a more detailed justification for employing the matching method.
Matching is justified by two assumptions:
1. The Conditional Independence Assumption (CIA):
(Y0, Y1) ⊥ D | X2. The Overlap assumption:
0 < Pr(D = 1 | X) < 1.
The first assumption denotes the statistical independence of (Y0, Y1) and D conditional on X.
Conditioning on those values, firms randomly select into bankruptcy status. CIA assumes that the
conditioning variables are sufficiently rich to justify the application of matching, and that the filing
decision does not systematically depend on unobservables.
The overlap assumption asserts that every individual in the population has a chance of receiving
the treatment. This requirement guarantees that matches can be found for all values of X.
Given these two assumptions, the ATT is identified and selection bias eliminated, since
11
E[Y1 − Y0 | D = 1
]
= Ex
[E [Y1 |X, D = 1]− E [Y0 |X, D = 1]
]
= Ex
[E [Y1 |X, D = 1]− E [Y0 |X, D = 0]
]
As argued in Section III, the best way to estimate the impact of bankruptcy filing is to construct a
control group that is similar, in terms of observable pre-filing firm attributes, to the bankruptcy filing
firms and mimics the “counterfactual” outcome for filing firms. This is precisely what the matching
method does. It explicitly controls for pre-filing firm attributes by comparing the outcome for a filing
firm with outcomes for comparable non-filing firms with similar characteristics. Matching overcomes
several shortcomings of existing empirical studies. First, it does not assume any functional form
for the outcome variables since the “counterfactual” outcome is estimated nonparametrically; this
is to be preferred since no underlying theory predicts a particular functional form of the outcome
variables. Second, the matching method explicitly controls for pre-filing firm characteristics so that
a comparison is only made between the filing firms and matched non-filing firms with similar pre-
filing characteristics. This separates the pure effect of the bankruptcy filing from the influence of
the firm’s pre-filing performance. A comparison between filing firms and matched non-filing firms
also separates the effect of the bankruptcy filing from those of other common economic factors.
A typical matching estimator takes the form,
M = 1N1
∑i∈I1
[Y1i −
∑j∈I0
W (i, j) · Y0j
]
where I1denotes the set of treatment units (filing firms), I0 the set of comparison units (non-filing
firms), and N1is the number of units in the set I1. The matches for each treatment unit i ∈ I1 are
constructed as a weighted average over the outcomes of comparison units, where W (i, j) is a weight
with∑
j∈I0W (i, j) = 1 and it depends on the distance between Xi and Xj . Different matching
methods differ in how they determine matches and in how they assign weights to the matches.
We use two particular types of estimators in checking the robustness of the empirical results.
(i) Nearest-neighbor Matching
This method assigns to each unit i a unit j closest to i in terms of covariates X. In the case of
ties, the nearest neighbor j is chosen by a random draw. Unit j may be reused for other matches.
For this estimator, the weighting scheme assigns all the weight to the single match:
W (i, j) =
1 if j ∈ {j ∈ I0 | minj ‖Xi −Xj‖}0 otherwise
A natural extension is to use the K-nearest neighbors, for positive integer K.
12
To determine the distance between Xi and Xj , this paper uses the Mahalanobis metric:
‖Xi −Xj‖ = (Xi −Xj)′ · Σ−1Xi· (Xi −Xj)
where Σ−1Xi
is the covariance matrix of the covariates from the treatment group. This metric has the
attractive property that it reduces differences in covariates within matched pairs in all directions.
The above traditional nearest-neighbors matching pairs each treatment unit i with only K
matches that have the closest pre-treatment characteristics X under the Mahalanobis metric. The
Kernel-based matching estimator (Heckman, Ichimura, and Todd 1997) uses all individual units in
the comparison group to construct matches for treatment unit i and assigns different weights to
comparison units that have different distances to unit i.
(ii) Kernel matching
This method constructs matches by forming weighted averages of the outcomes of all units in
the D = 0 comparison sample. If weights from a typical symmetric, nonnegative, unimodal kernel
are used, then the average places higher weight on units that are close in terms of X and a lower
weight on more distant observations. Kernel matching defines W (i, j) = Kij∑j∈I0
Kij,
where Kij = K((Xi−Xj)/λ) is a kernel function and λ is a bandwidth parameter. Under standard
conditions on the bandwidth and kernel,∑
j∈I0W (i, j)·Y0j =
∑j∈I0
Y0j ·Kij∑j∈I0
Kijis a consistent estimator
of E(Y0| D = 0, Xi), which is equal to E(Y0| D = 1, Xi) under the CIA assumption.
In practice, matching may be difficult to implement when the set of variables in X is large. For
example, kernel matching estimates non-parametrically the conditional mean E(Y0|D = 1, X) and
if the dimension of X is large, the convergence rate will be slow due to the “curse of dimensionality”
problem. Rosenbaum and Rubin (1983) derive an important result that potentially solves this
problem. Let Pr(D = 1|X) , P (X); this is called the propensity score, the probability of receiving
the treatment conditional on covariates X. Rosenbaum and Rubin demonstrate that the CIA and
Overlap assumptions together imply that (Y0, Y1) ⊥ D | P (X) and 0 < Pr(D = 1|P (X)) < 1. This
means that matching can be performed on the propensity score P (X) alone, reducing a potentially
high-dimensional matching problem to a one dimensional problem. The above discussion about
matching X carries over to P (X); Kernel matching can thus be implemented based on P (X).
Another practical issue concerning the matching method is the so-called common support prob-
lem, which is the empirical counterpart to the overlap assumption, which states that every individual
in the population has a chance of receiving the treatment. That is, for every value of X (or equiv-
13
alently P (X)) there are both treatment units and comparison units. Empirically, this assumption
requires the density functions of P (X) to have common support regions for both the treatment
and comparison groups. This paper uses a method proposed by Heckman et al. (1997) to impose
a common support when implementing matching on the propensity score. Details of the procedure
are contained in Appendix B.
Before turning to the estimation results we discuss in the next section bankruptcy filing deter-
minants that are used in our matching model.
V. Determinants of Bankruptcy Filing and Empirical Measures
To implement the matching method, determinants of bankruptcy filing, the observable covariates
X, need to be identified. The CIA condition is more likely to hold if the filing determinants that
are identified provide sufficiently rich information so that any uncontrolled or unobserved factors
do not systematically affect the filing decision.
This section does not provide a survey of why firms default and how to resolve default. Instead
we rely on the extant literature to present a simplified framework to support the empirical analysis
and to identify factors that potentially affect firms’ bankruptcy filing decisions.2
A firm’s bankruptcy decision can be decomposed into two steps, whether to default and, if so,
how to resolve the problem of default. At any point in time, the firm faces the options of either
continuing with its current operations (for example, it may want to raise external funds if it is
short of cash to pay off current debt) or defaulting on its current debt obligations. If it defaults,
the firm can either reorganize through a private workout or a formal bankruptcy filing, or it can
liquidate its assets. The equity holders (if they are in control of the firm) make this decision based
on equity value in three alternative situations, namely, continuation (Ec), reorganization (Er), and
liquidation (El).
Suppose equity value is Ec if the firm continues its current operation and, in addition, that
current equity value can be affected by the firm’s liquidity. If the firm lacks sufficient cash reserves
or slack to cover its current debt obligations and normal operations, it needs to go to the external
capital market to raise funds. In a capital market with transaction costs and asymmetric informa-2The Accounting Literature attempts to find empirically factors (e.g., financial ratios) that have explanatory
power in predicting bankruptcy, but it lacks an underlying theory. The Game Theory literature on financial distress,on the other hand, offers theoretical explanations on how the distressed firm chose (or should choose) alternativemeans of resolving financial distress. It is worth integrating the above two literatures, as we try to do.
14
tion, where firm insiders know more about the firm’s future prospects than the capital market at
large, the marginal cost of outside financing tends to exceed that of inside financing. As shown
by Myers and Majluf (1984), in such a world with rational expectations, ex ante shareholders’
equity value will be lower than in a frictionless capital market with complete information. Thus,
outside financing could be costly to the firm’s shareholders, since it reduces equity value under
continuation, Ec. Taking this into account, the firm’s shareholders can compare Ec with Er and
El to decide whether or not to default. These values are also a function of the possibility of rene-
gotiation between shareholders and debtholders. Whether such renegotiations take the form of a
private workout or occur through a formal bankruptcy filing depends on the transaction costs of
bargaining. As discussed in Section II, the more severe are these bargaining problems, the more
likely it is that the firm will resort to Chapter 11 filing.
[ Insert Table I About Here ]
Such considerations suggest a set of firm and market attributes as determinants of the firm’s
filing decision. Summarized in Table I and listed below are corresponding empirical measures that
are constructed from COMPUSTAT and used as determinants of bankruptcy filing.
(i) Equity value:
As discussed, equity holders make their default and renegotiation decision in part by comparing
potential equity values under continuation (Ec), reorganization (Er), or liquidation (El). The larger
is Ec relative to the other two, the less likely is the firm to default and file for bankruptcy. As an
empirical proxy for Ec, the market value of equity, normalized by the total liabilities, is used in this
paper.
(ii) Liquidity:
A firm in financial difficulty may seek outside financing in order to continue. In a world of
asymmetric information, outside financing tends to be costly and reduces current shareholders’
equity value; this makes financial slack valuable to the firm. The more financial slack the firm has,
the less likely it is to need outside financing and this attenuates the reduction in equity value and
reduces the likelihood of default. Two empirical measures are used as proxies for the firm’s liquidity
level. The first is the firm’s quick ratio defined as cash plus receivables divided by current liability,
while the second is the interest coverage ratio, measured as EBIT over current interest expenses.
15
The latter captures the firm’s ability of generating sufficient revenues to meet interest expenses.
The firm’s ability to raise outside funds is also affected by its fixed assets. A firm with more
fixed assets that can serve as collateral is better able to borrow new funds to overcome default. As
a proxy for fixed assets, this paper uses the firm’s total property, plant and equipment, normalized
by the total liability.
(iii) Profitability:
Financial difficulties cause the firm to resort to outside financing, reducing shareholders’ equity
value. The most important determinant of financial difficulty is the firm’s profitability, and two
profitability measures are used: (1) the ratio of sales to total assets, and (2) the ratio of cash flow to
total assets. These are asset-turnover ratios illustrating the sales or cash-flow generating potential
of the firm’s assets.
(iv) Liquidation value:
The lower is the liquidation value of the firm, the more likely it is to choose reorganization
rather than liquidation. To capture this notion, this paper uses the firm’s total intangible assets
normalized by total assets; a firm with a greater proportion of intangible assets tends to have a
lower value in the event of liquidation, and is more likely to be reorganized rather than liquidated.
Moreover, as Gilson, John, and Lang (1990) argue, its assets are more likely to be sold when debt
is restructured under Chapter 11 rather than privately. Chapter 11 is more costly for firms with a
higher proportion of intangible or firm-specific assets, and thus, such firms are less likely to file for
formal bankruptcy.
(v) Measures Reflecting Bargaining and Recontracting Problems:
A firm’s choice between a private workout and formal bankruptcy filing depends on the severity
of bargaining problems; bargaining and recontracting problems make it less likely to achieve a
successful private workout. Two types of bargaining problems can be particularly serious, the
holdout problem and the asymmetrical information problem. The severity of the holdout problem
is affected by the heterogeneity of the debt claims, since greater heterogeneity implies greater
bargaining difficulties in having debt restructured. We consider two types of debt claims, the trade
credit ratio and the secured debt ratio. As discussed in Gilson et al. (1990), the holdout problem is
particularly severe for trade credit debt because the number of trade creditors is often quite large,
16
and their claims are relatively heterogeneous.3 Achieving a consensus through bargaining among
trade creditors also tends to be more difficult. We use accounts payable normalized by total liability
as the indicator of the trade credit ratio, where a larger trade credit ratio tends to generate a more
severe holdout problem in private bargaining leading to a greater probability of formal bankruptcy
filing. Secured debt, on the other hand, tends to be more homogenous since it is usually purchased by
a relatively small number of institutional investors, e.g., banks and insurance companies. Moreover,
institutional investors tend to have greater incentives and expertise to monitor a firm’s operation,
and thus face less severe information problems. These considerations imply that a firm with more
secured debt tends to rely relatively more on private workout in a reorganization. Note, however,
that the holdout problem could also be affected by the creditor’s bargaining position in a private
workout and in this regard, secured creditors are less likely to tender their claims in a private
workout since their claims are secured by the pledge. Thus, the impact of the secured debt ratio
on the firm’s reorganization choice is ambiguous. The empirical proxy we employ for this ratio is
the firm’s mortgage and other secured debt normalized by its total liability.
Finally, we follow Chen (2003) in using the auditor’s opinion as an empirical proxy for asym-
metric information problems. There are six categories for the Auditor’s Opinion item in the
COMPUSTAT database and we assign a score of 1 to 4 (the higher the score, the better is the
quality of the disclosure of information) as follows:
1. There is either (a) an adverse opinion where an auditor has expressed a negative opinion on
the financial statements of the company, or (b) a qualified opinion where the financial statement
reflects the effects of some limitation on the scope of the examination or unsatisfactory presentation
of financial information.4
2. Either financial statements are unaudited, or there is no opinion, where the auditor refuses
to express an opinion regarding the company’s ability to sustain its operation as a going concern.
3. Unqualified Opinion with Explanatory Language, where the auditor has expressed an un-
qualified opinion regarding the financial statements but has added explanatory language to the
auditor’s standard report.
4. Unqualified Opinion, where the financial statements reflect no unsolvable restrictions and3A large number of claimants exacerbate free-rider and empty core problems in bargaining.4these two categories correspond to data codes 5 and 2 respectively. This paper groups these two categories into
one, because (1) in the sample only 13 observations have a code value of 5, and (2) code 5 is only available from 1988forward.
17
the auditor makes no significant exceptions as to the accounting principles, the consistency of their
application, and the adequacy of the information disclosed.
The above 10 firm attributes constitute the pre-filing covariates X employed in this study. The
CIA condition of the matching method assumes that a sufficiently rich set of factors have been
identified so that any uncontrolled or unobservable factors do not systematically affect firms’ filing
decisions.5
In conditioning on the 10 firm attributes, we construct a control group as a benchmark to
compare with the filing firms. We believe that firm heterogeneity is the central issue that should be
taken into account in assessing the impact of bankruptcy filing on firm performance. We move from
a unidimensional industry median control variable used in extant models to firm-specific attributes.
We do not explicitly control for industry classification, and it is possible that a filing firm is matched
with a non-filing counterpart in a different industry; but this is not as problematic as it appears.
First, we do not think that the industry classification code is a good determinant of bankruptcy-
filing. But even if it is, the 10 firm attributes will pick up, to some extent, the industry effect
in filing decisions. Second, comparing a filing firm with non-filing firms outside the filing firm’s
industry may not be a big problem if there are similarities in characteristics such as profitability
and liquidity, and moreover common economic factors tend to affect operating performances of
similar firms in similar fashion. If there were industry-specific shocks, the above two points would
be weakened but not substantially.
Note that if the industry-specific shock is purely random, it does not systematically affect
firms’ filing decisions conditional on controlled firm attributes. When industry classification is not
explicitly controlled for, it is implicitly grouped in the set of unobservable factors. This does not
bias the average effect of bankruptcy filings (the ATT parameter estimated in this paper), since the
industry classification would not affect filing decisions systematically and the CIA condition would5Another factor which is worth investigating is the pension issue that has been an important factor affecting firms’
Chapter 11 filing decisions in the past few years. This paper considers two measures to investigate the influence ofpensions. One measure is the unfunded pension liability. This is equal to the actuarial present value of accumulatedpension benefit obligation (COMPUSTAT data item 285) minus pension plan assets (COMPUSTAT data item 287).The second measure is pension and retirement expenses (COMPUSTAT data item 43). Unfortunately, the above twomeasures generated from COMPUSTAT are not sufficiently informative. Specifically, beside the previously identified10 factors, controlling for the two measures drops eligible filing firms by proportions of 1/2 and 1/5, respectively.Furthermore, it is not clear whether the missing observations are due to the fact that firms are not reporting or tothe fact that they do not have pension plans. Lastly, since the sample coverage is from 1980 to 2003 in this paper,it is not clear whether the pension issue is a prominent one for the entire sample period. Based on above reasons,we do not control for pension factors and the analysis and results in what follows are based solely on the previouslyidentified 10 factors.
18
still hold to justify the matching method.
However, if firms can perfectly foresee industry-specific shocks then, conditional on similar firm
attributes, filing decisions will differ systematically across industries. In this case, not controlling
for industry classification leads to a violation of the CIA condition. The extent to which this is
problematic depends on how much foresight firms have. In general, in this situation the matching
estimator is biased as uncontrolled or unobservable factors lead to a violation of the CIA condition.
As shown in Appendix A, however, in this case, matching methods can at least reduce the estimation
bias by eliminating the biases due to the other two sources, the non-overlapping support and the
difference in the distributions of the observable attributes X between treatment and comparison
groups.
Finally, even when industry classification is an issue, explicitly controlling for it may not atten-
uate the estimation problem, but instead lead to biased results. Suppose we restrict firms to the
same industry and then match them conditional on the 10 characteristics. This may also create
bias in the estimates because the operating performances of similar firms in the same industry tend
to be correlated, e.g., one firm’s losses may contribute to gains to its competitors.
VI. Empirical Results
A. Data Description
The paper estimates the impact of Chapter 11 bankruptcy filings on U.S. public companies.
Information on the sample of public U.S. bankrupt companies is from the Bankruptcy Research
Database (BRD),6 kindly provided by Professor Lynn M. LoPucki. This database includes Chapter
11 bankruptcy cases filed since 1980, by or against a company, that (1) has assets worth $100 million
or more at the time of filing (measured in 1980 dollars), and (2) is required to file 10-Ks with the
Securities and Exchange Commission (SEC). Primary information extracted from this database
is the date of the Chapter 11 bankruptcy filing, and the date of a reorganization plan confirmed
by the bankruptcy court. To get balance sheet information, the companies in BRD are linked to
COMPUSTAT via the unique identifier GVKEY in the latter database.
Initially, we have 667 Chapter 11 bankruptcy filings between 1980 and 2003. Among them, 619
cases in which the companies involved can be identified in COMPUSTAT. Table II provides detailed6A web-based version of BRD is available at http://lopucki.law.ucla.edu/index.htm
19
summary statistics across years for these 619 cases. It reports the total number of filing cases, the
number of cases in different outcome categories, and the time spent in Chapter 11 for those that
eventually emerge from bankruptcy. For the entire sample, 369 filing cases emerge from Chapter
11 bankruptcy, while 208 cases do not; 34 cases were still pending until 2005 and the remaining 8
cases lack data to verify their status. The term “emerge” refers to the plan of reorganization by
a surviving firm to be confirmed and consummated. An emerging firm is a stand-alone firm that
continues to operate after confirmation of the reorganization plan. In the following cases no firms
emerge: (1) if the firm’s assets are integrated into the business of the acquirer or merger partner
during bankruptcy or under the reorganization plan; (2) the firm is liquidated during bankruptcy;
(3) a firm that plans to liquidate pursuant to the plan does not emerge, even if it continues to
operate after confirmation and consummation. For the firms that emerged from bankruptcy, it
took an average of 16.22 months to emerge from bankruptcy. The minimum was 1 month and the
maximum 83 months.
[ Insert Table II About Here ]
A further breakdown of the 208 non-emerged cases shows the following:
(1) 62 cases belong to a “§363 sale”. A“§363 sale” means that the debtor sold all or substantially
all of its assets during the Chapter 11 case. Thereafter, the court may have confirmed a plan
distributing the proceeds of the sale (“§363 sale confirmed”) or converted the case to Chapter 7
(“§363 sale converted”);
(2) 24 of them were formally converted to chapter 7 proceedings;
(3) 120 of them had their plan confirmed. For these firms, most of them were liquidated at confir-
mation and some of them were acquired;
(4) 2 cases were dismissed.
For the 369 emerging cases, we have post-bankruptcy sales information for 213 cases. We do
not have complete information in COMPUSTAT for the remaining 156 cases because some firms
went private after emerging, some did not file data with the SEC, and some were deleted from
COMPUSTAT because of acquisition and merger.
20
B. Estimation Results
A firm’s bankruptcy filing decision is referred to as a treatment. A filing firm is called a treatment
unit and all treatment units constitute the treatment group. Similarly, non-filing firms are called
comparison units and comprise the comparison group. The parameter of interest is the Average
Treatment effect on the Treated (ATT). The central question is whether filing firms perform, on
average, at least as well as they would have had if they had not filed for bankruptcy. To investigate
the research question, the following hypothesis is made:
H0 : ATT = 0 and H1 : ATT > 0.
To estimate ATT, the matching estimator M is used. In particular, two types of operating
measures are used to calculate matching estimates: sales and cash flows.
The treatment group is extracted from BRD, including 213 Chapter 11 bankruptcy cases that
eventually emerged and had sales information for at least one year after emergence. The comparison
group is retrieved from the entire dataset in COMPUSTAT, excluding firms identified in LoPucki’s
bankruptcy dataset or firms that became inactive in COMPUSTAT due to Chapter 11 or Chapter 7
bankruptcy. Furthermore, we focus on large firms and require that comparison units be large firms,
since the BRD database contains only large corporations. Specifically, we require a comparison firm
to have assets worth more than $100 million, measured in 1980 dollars, for more than 80% of all
firm-years for which it has asset value information available.7 In order to get a firm’s information
before filing and after emergence, the accounting variables for both groups, from 1978 to 2004, were
retrieved from COMPUSTAT. Finally, all current-year dollars are deflated to 1980 dollars.
For a treatment unit, the timing is as follows. Denote by t0 the year in which a firm files for
bankruptcy and t1 as the year when it has its plan confirmed and emerges from bankruptcy. Denote
by t−1 the pre-filing year, the first year before filing for which relevant data are available; t2 denotes
the post-emergence year, the first year after emergence for which outcome variable data are available.
The 10 bankruptcy determinants identified in Section V, denoted by vector X, are the pre-filing
characteristics for each unit. The outcome measures under consideration are sales and cash flows,
normalized by total assets, in the post-emergence year. It is required that the treatment units have7The incompleteness of historical SIC information in COMPUSTAT does not allow us to accurately track a firm’s
SIC over the sample period, 1980-2003. So in the results reported below, we do not further eliminate financialinstitutions. However, as an expedient, we make use of the current SIC information in COMPUSTAT (variable dataitem DNUM) and find that 7 of the filing firms are financial institutions (SIC starting with 6). We eliminate those7 filing firms and all financial firms in the comparison group, and our results below do not change.
21
data available for the full set of covariates X in the pre-filing year t−1. Elimination of cases with
incomplete information eventually results in 143 treatment units.8 The summary statistics of the
pre-filing characteristics for both the treatment and comparison groups are reported in Table III.
Based on the pre-filing characteristics X, the traditional K-nearest-neighbor matching method
was implemented as follows: Let i denote a treatment unit and j a comparison unit. For each unit i,
its matched comparison units are chosen by the following steps: (1) Extract the covariates of unit i in
its pre-filing year, Xi; (2) for each unit j, extract its covariates in the same year as unit i′s pre-filing
year; (3) calculate the distance between i and j in terms of X; to determine the distance between
units Xi and Xj , the Mahalanobis metric is employed, i.e., ‖Xi −Xj‖ = (Xi−Xj)′ ·Σ−1Xi·(Xi−Xj);
(4) the k nearest j units are chosen as matches for unit i; the simple average is then taken for the
k matched comparison units’ outcome measures for the same period as treatment unit i′s post-
emergence year; the average is the estimated counterfactual outcome for the treatment unit i; (5)
after all treatment units are matched, the difference between the means of the treatment units’
observed outcomes and their estimated counterfactual outcomes is computed as the estimate of the
Average Treatment effect on the Treated.
Table IV shows the results for k-nearest-neighbor matching with k=1, 4, 6, 10. Two measures
are considered in Table IV: sales and cash flows, both divided by total assets. The two outcome
measures are normalized by total assets as they reflect the sales and cash flow generating ability of
the firms’ assets. These unit-free variables are robust to firms’ asset shrinkage during bankruptcy.9
In Panel A, the outcome variable is the value of each of the two measures in the post-emergence
year of a filing firm. In Panel B, the outcome variable is the difference between the values of each
performance measure in the post-emergence year and in the pre-filing year. That is, the outcome
variable reported in Panel B is the performance measure’s value change from the pre-filing year to
the post-emergence year. In both panels, results are reported for k-nearest-neighbor matching. The
results are quite similar for different nearest-neighbor matching.
[ Insert Table IV About Here ]
8For the 143 treatment units, post-emergence years t2 are either 1 or 2 years after confirmation years t1; mostpre-filing years t−1 are 1 or 2 years before filing years t0. But for 22 cases, pre-filing years are more than 2 yearsbefore filing. In order to avoid losing additional treatment units, those cases are not deleted. However, we alsocarried out our empirical work on the 121 cases that have pre-filing data available no longer than 2 years beforefilings. The results are qualitatively the same.
9Indeed, we find that, for the 160 treatment units, the median asset value is 334.47 million dollars in the pre-filingyear, 197.33 million dollars in the post-emergence year.
22
Take as an example the 10-nearest-neighbor matching for the sales-to-total-assets ratio in Panel
A. The average outcome for the 143 treatment units is 1.447 while the average outcome of all the
matched comparison units is 1.149, implying a positive Average Treatment Effect on the Treated
(ATT) of 0.298. This means that filing firms’ sales-to-total-assets ratios are, on average, 29.8
percentage points higher than what they would have been if the firms had not filed for bankruptcy.
Thus the sales-generating ability of firms’ assets improves if they file for bankruptcy. To assess
statistical significance, the bootstrap distribution and rejection regions are generated for the t-
statistic based on 1000 bootstrap replications. The result for the 10-nearest-neighbor matching
shows that we can reject H0 in favor of H1, that ATT is positive at the 1% significance level. In
terms of the cash flow measure, the ATT results are all negative but the magnitudes are small, and
none of them is statistically different from zero at the 10% level.
In Panel B, the outcome variables are interpreted as changes in firm performance from before to
after bankruptcy. Take the 10-nearest-neighbor matching case as an example. On average, the filing
firms’ sales ratios increase by 0.235 from the pre-filing year to the post-emergence year, while that
change is 0.041 for matched comparison firms. This implies a positive ATT of 0.193, which means
that the sales ratio increases 19.3 percentage points more compared to matched units not filing
for bankruptcy. This result is statistically positive at the 1% significance level. When the change
in cash flow outcome from before to after bankruptcy is considered, the ATT results become all
positive although not all of them are statistically positive. Take again 10-nearest-neighbor matching
as an example. On average, filing firms’ cash-flow ratios increase by 0.168 from the pre-filing year
to the post-emergence year, while the ratios for the matched comparison firms increase by 0.033.
This implies a positive ATT of 0.135, which is statistically positive at the 5% level.
Different nearest-neighbor matching estimates for the sales ratio and cash flow ratio measures
all imply positive Average Treatment effects on the Treated, except for the cash flow measure
at the post-emergence year. Most positive results are significant at conventional levels. The few
negative results for cash flow over assets have small magnitudes and are not statistically different
from zero. Overall, these findings provide evidence that Chapter 11 filings improve firms’ operating
performance, compared to their performance without filing.
In addition to the traditional matching method, this paper also employs the propensity score
matching method to estimate the ATT. Ideally, a multinomial logistic regression needs to be used to
23
estimate the propensity to file for formal bankruptcy. But our data do not allow us to distinguish
other categories of the dependent variable such as whether a firm was reorganized informally.
Instead, we run a simple binary logistic regression, with the dependent variable equal to 1 if the
firm files for bankruptcy and equal to 0 if the firm does not do so for whatever reason. The
explanatory variables are the ten filing determinants discussed in Section V. The purpose is to
obtain a unidimensional index, the probability of filing. Based on this, the propensity score matching
exercise is carried out and used to supplement the traditional matching results based on the same
10 firm attributes. The results of the logistic regression are shown in Table V.
[ Insert Table V About Here ]
Tables VI and VII present results for propensity score matching. Based on the estimated
propensity scores, both K-nearest-neighbor and kernel-based matching are implemented. The K-
nearest-neighbor matching is implemented in the same way as before, except that matching is now
based on propensity scores instead of the covariate X. The kernel-based matching is similarly
implemented except that for each treatment unit, the counterfactual outcome is constructed as a
weighted average of all comparison units with weights being assigned by a normal kernel function.
In Table VII, the common support region is imposed using the method discussed in Appendix B,
with a trimming rule q=0.01.
The results in tables VI and VII are generally consistent with the findings in Table IV for the
traditional matching method. Matching estimates for the sales ratio and the cash flow ratio show
positive Average Treatment effects on the Treated, except for the cash flow measure in the post-
emergence year. Again, the negative results have small magnitudes and are not statistically different
from zero at the 10% significance level. For the positive results, most of them are statistically
positive at conventional levels for the sales measure. When the cash flow measure is considered,
the positive results are smaller and not significant compared to the results in Table IV. Overall,
the estimation results provide evidence that bankruptcy filings have benefits for the successfully-
emerged large public companies considered in our sample.
[ Insert Tables VI and VII About Here ]
The rest of the discussion in this sub-section investigates the robustness of our results. First, we
24
divide our original sample into two parts, using 1990 as a cut-off filing year. Matching exercises are
repeated for the sub-sample of filing cases after 1990. In this sub-sample, we have 114 cases that
have all relevant data information and are used for matching. The results for sales-to-assets ratio
and cash-flow-to-assets ratio are qualitatively the same as those for the entire sample, and they
are reported in Table VIII. For majority of the positive estimates, they are statistically positive
at conventional levels. For the few negative estimates, their magnitudes are small and are not
statistically different from zero at the 10% level.
[ Insert Table VIII About Here ]
Second, we split the sample according to the duration of bankruptcy for the emerged cases. In
the matching sample of 143 filing firms, the longest time spent in bankruptcy is 70 months. We
are interested in how the duration in bankruptcy affects existing results. In particular, the entire
sample is divided into two sub-samples whose filing firms stay in bankruptcy for less than and more
than 1 year, respectively. We have 71 cases that emerge within 1 year, and 72 cases that stay in
bankruptcy for more than 1 year. The results are presented in tables IX and X. Compared with our
earlier results, the major patterns remain. When comparing results of the more-than-1-year sample
with those of the less-than-1-year sample, we find that the results for the sales measures are similar
in the two sub-samples, and both are positive. In terms of the cash flow measures, the estimates
for the less-than-1-year sample are now positive in the post-emergence year, although they are not
statistically positive. In contrast, the results are negative in the more-than-1-year sample and most
are statistically negative (not reported in the table). Only in this case, do we have some evidence
that firms staying in bankruptcy longer fare worse compared to those staying for a shorter period.
[ Insert Tables IX and X About Here ]
Finally, we repeat the matching exercises conditioning on firm attributes at different pre-filing
years. In our above results, the pre-filing year, t−1, is the first year before the filing year for
which data are available. One could argue that if firms rationally anticipated their bankruptcy
filings and made adjustments accordingly, then matching right before the filing year may not be
appropriate and may not capture the real impact of the bankruptcy filing. Thus we chose different
pre-filing years, t−1, to see whether our results are changed. In particular, we repeated the matching
25
estimations with the pre-filing year as the second and the third year before filings, respectively. Our
earlier results did not change qualitatively.10
C. Discussion of the Survivorship Bias
There are 619 Chapter 11 bankruptcy cases in the initial sample. Of these, 208 cases do not
emerge while 369 cases do. Among the latter, we have 156 cases for which no post-bankruptcy
information is available for various reasons: some firms went private after emerging, some did not
file fundamental data at the SEC, some were deleted from COMPUSTAT because of acquisition
and merger.
The non-emerged cases involve either 1) the sale of all, or substantially all, of the firm’s assets,
or 2) liquidation at the end of chapter 11. Lack of data for these cases leads to a survivorship bias,
but this is not inconsistent with our finding that chapter 11 tends to improve economic efficiency in
the sense that the emerged firms, on average, performed better than they would have had they not
filed for bankruptcy. It is certainly the case that the non-emerged firms were undergoing sale or
liquidation. To evaluate fully the impact of Chapter 11, the correct question is how would have the
non-emerged firms performed without Chapter 11 filings. Presumably, these firms failed to propose
viable reorganization plans, and some of them would be liquidated without Chapter 11 filings.
For the 213 emerged cases with post-bankruptcy information, we do find that their average asset
size shrunk after emergence. It is difficult to empirically track all the assets sold during Chapter 11.
Conceptually, however, ignorance of the asset sales is not inconsistent with our empirical findings
as long as the assets sold elsewhere were put to better use.11
Regarding the 156 emerged cases that lack post-bankruptcy information, we cannot say more
about them given the absence of reasons why those firms chose to go private after emergence.
VII. Conclusion
This paper investigated the impact of Chapter 11 bankruptcy filing on U.S. public firms. In
the presence of bargaining and information problems, private workouts often fail to induce efficient
outcomes, which may justify why many firms file for Chapter 11 bankruptcy. The research question10To save space, these results are not reported but are available upon request.11In Maksimovic and Phillips (1998), they find evidence using plant-level data for manufacturing firms, that plants
have higher Total Factor Productivity when sold to new owners post chapter 11.
26
investigated in this paper was whether Chapter 11 bankruptcy filing improved, on average, firms’
operating performance compared to their performance without such filing. To empirically evaluate
the impact of bankruptcy filing, the estimation strategy we used entailed controlling for firms’
pre-filing characteristics. This served to isolate the pure effect of bankruptcy filing. The paper
used matching methods to estimate the impact of Chapter 11 bankruptcy filing on firms’ operating
performance. This method helped control for firm attributes and dealt with problems engendered
by firm-level heterogeneity. Both traditional matching and propensity score matching methods were
implemented, and the results obtained were quite consistent. Overall, they showed positive impacts
of chapter 11 bankruptcy filings on the filing firms’ performance, measured in terms of sales or cash
flow divided by total assets. The results were robust with respect to different filing periods, and also
were essentially unchanged when matching was conditioned on firm attributes at different pre-filing
years. Overall, the results indicated that chapter 11 bankruptcy filings were beneficial to the large
public firms in this sample that successfully emerged from bankruptcy. Note that our results do not
show that formal bankruptcy filing is always a better mechanism than alternatives such as informal
reorganization. Instead, the question of interest addressed in the paper was whether firms that
filed under chapter 11 performed better after bankruptcy than they would have without filing. Our
results indicate that formal bankruptcy filing helps such firms.
27
Appendices
A. Matching Methodology
Matching methods are widely used in the public policy and social program evaluation literature.
We provide a brief introduction to the matching method and relate its application to evaluating the
impact of the bankruptcy filing. The following discussion draws on Heckman et al.(1997, 1998b).
In the context of bankruptcy filing, a firm’s filing decision is labeled as a treatment. A filing
firm is called a treatment unit in the treatment group with Di = 1. A non-filing firm is called
a comparison unit in the comparison group with Di = 0. Each firm has a vector of pre-filing
characteristics referred to as covariates and denoted by Xi. The operating outcomes for a filing and
a non-filing firm are denoted by Yi1 and Yi0 respectively. The outcomes used in this paper relate
to the firm’s operating performance such as sales and cash flows.
To evaluate the impact of bankruptcy filings, the parameter of interest is given by E[Y1−Y0|D =
1] = 1,the Average Treatment effect on the Treated (ATT). It captures the (average) difference
between the effect on the units being treated and the effect as if they had not received the treatment.
The mean E[Y1|X, D = 1] can be identified from data on filing firms. The mean E[Y0|X,D = 1] is
called the counterfactual mean and it cannot be identified from data.
If social experiments were available that assigned firms bankruptcy status randomly, the random
assignment would ensure that the filing firms and the randomized-out non-filing firms have the same
characteristics, both in terms of observed and unobserved features. Social experiments provide
information needed to form the sample counterpart of E[Y0|X, D = 1] and hence to provide an
unbiased estimator of the ATT parameter.
But observational data on bankruptcies do not provide the sample counterpart of E[Y0|X,D =
1].To estimate the ATT, it is intuitive to replace E[Y0|X, D = 1] with E[Y0|X, D = 0] for which
data are available. However, a naive comparison between E[Y1|X,D = 1] and E[Y0|X, D = 0] leads
to the so called selection bias for observational studies.
E [Y1 − Y0 | D = 1]
= Ex
[E[Y1 − Y0 | X,D = 1]
]
= Ex
[E[Y1 | X, D = 1]− E[Y0 | X, D = 0] + E[Y0 | X,D = 0]− E[Y0 | X,D = 1]
]
The bias is B = Ex
[E[Y0 | X, D = 1]− E[Y0 | X, D = 0]
]= E[Y0 | D = 1]− E[Y0 | D = 0].
28
Matching is a way to estimate the evaluation parameter using observational data. It reduces
selection bias by constructing a control group of “comparable” units. In particular, it is justified
by two assumptions.
1. The Conditional Independence Assumption (CIA)
(Y0, Y1) ⊥ D | X,
which denotes the statistical independence of (Y0, Y1) and D conditional on X.
2. Overlap assumption
0 < Pr(D = 1 | X) < 1
This requirement guarantees that matches can be found for all values of X.
Given the two assumptions, the ATT is identified, since
E[Y1 − Y0 | D = 1
]
= Ex
[E [Y1 |X, D = 1]− E [Y0 |X, D = 1]
]
= Ex
[E [Y1 |X, D = 1]− E [Y0 |X, D = 0]
]
Matching emulates some of the features of random experiments, (a) by aligning the distribution
of observed characteristics in the D = 0 group with that in the D = 1 group matching ensures
that the distribution of observed characteristics are the same among the D = 1 and D = 0 groups;
(b) CIA replaces randomization with conditioning on a set of X variables, or put differently, the
CIA assumes that the conditioning variables available to the analyst are sufficiently rich to jus-
tify the application of matching, and that the filing decision does not depend systematically on
unobservables.
Under the assumptions of the matching method, selection bias is eliminated. But although
matching cannot eliminate all selection bias, when the CIA is violated it does reduce it. Heckman
et al. (1997) decompose this bias into three parts, B1, B2, B3:
(i) B1 arises because of nonoverlapping support of X between D = 1 and D = 0 groups. For
some treatment units there are no comparable comparison units and for some comparison units there
are no comparable treatment units. Failure to compare at common values of matching variables
causes mismatching bias.
(ii) B2 arises from different distributions of X within the two populations. Failure to weight the
two groups comparably leads to misweighting bias.
(iii)B3 is due to differences in outcomes that remain even after conditioning on observables and
29
making comparisons on a region of common support; it is due to selection on unobservables.
They then show that matching eliminates two of the three sources of selection bias. The bias due
to non-overlapping supports is eliminated by matching only over the region of common support.
The bias due to different density weighting is eliminated because matching methods effectively
reweigh the comparison units. So only the bias due to differences in unobservables across groups
remains, and this occurs when we do not have sufficiently rich data to satisfy the CIA condition.
As discussed in Section III, only controlling for industry median does not isolate the pure
effect of bankruptcy filing. Estimation of the impact of bankruptcy filing on firms’ performance
requires constructing an appropriate control group to compare with. Ideally, the control group
would best mimic the “counterfactual” outcome of the filing firm. So the question is not whether
this comparison should be made, but rather how to find better comparison units to reduce bias.
This is what the matching method does, and the advantages of the matching method, as discussed
in Section IV, can improve estimation accuracy.
B. Algorithm for Generating Common Support Regions
This appendix outlines the algorithm in Heckman et al. (1997) that imposes the common
support region for propensity score matching.
By definition, the region of common support, S10, includes only those values of P that have
positive density within both the D = 1 and D = 0 groups. To determine this region, Heckman et
al. (1997) first estimate the densities at all the sample P values using a kernel density estimator
and determine S10 by forming S10 ={
P ∈ S1 ∩ S0 : f(P | D=1) > 0 and f(P | D=0) > 0}
,
where S1 and S0 are the estimated smoothed supports for the kernel density functions. Then they
exclude an additional q percent of the P points for which the estimated density is positive but very
low. The set of points potentially eligible for matches are given by
Sq ={
P ∈ S10 : f(P | D=1) > Cq and f(P | D=0) > Cq
},
where Cq satisfies sup
{Cq : 1
2J
∑I1
[1
(f(P | D=1) < Cq
)+ 1
(f(P | D=0) < Cq
)]≤ q
}
where I1 is the set of observed values of P that lie in S10 and J is the number of elements in I1. q
is called the trimming rule.
30
REFERENCES
Aivazian, Varouj V., and Jeffrey L. Callen, 1980, Corporate leverage and growth: The game-
theoretical issues, Journal of Financial Economics 8, 379–399.
Aivazian, Varouj V., and Jeffrey L. Callen, 1983, Reorganization in bankruptcy and the issue of
strategic risk, Journal of Banking and Finance 7, 119–133.
Altman, Edward I., 1984, A further empirical investigation of the bankruptcy costs question, Jour-
nal of Finance 39, 1067–1089.
Altman, Edward I., 2000, Predicting financial distress of companies: Revisiting the z-score and
zetaR© models, Working Paper, New York University.
Andrade, Gregor, and Steven N. Kaplan, 1998, How costly is financial (not economic) distress?
evidence from highly leveraged transactions that became distressed, Journal of Finance 53,
1443–1493.
Brown, David T., 1989, Claimholders incentive conflicts in reorganization: The role of bankruptcy
law, Review of Financial Studies 2, 109–123.
Chen, Nan, 2003, An empirical study of a firm’s debt restructuring choices: Chapter 11 vs. workouts,
Working Paper, Columbia University.
Giammarino, Ronald M., 1989, The resolution of financial distress, Review of Financial Studies 2,
25–47.
Gilson, Stuart C., Kose John, and Larry H.P. Lang, 1990, Troubled debt restructurings: an empirical
study of private reorganization of firms in default, Journal of Financial Economics 27, 315–353.
Haugen, Robert A., and Lemma W. Senbet, 1978, The insignificance of bankruptcy costs to the
theory of optimal capital sturcture, Journal of Finance 33, 383–393.
Haugen, Robert A., and Lemma W. Senbet, 1988, Bankruptcy and agency costs: Their significance
to the theory of optimal capital sturcture, Journal of Financial and Quantitative Analysis 23,
27–38.
31
Heckman, James J., Hidehiko Ichimura, Jeffrey Smith, and Petra E. Todd, 1998a, Characterizing
selection bias using experimental data, Econometrica 66, 1017–1098.
Heckman, James J., Hidehiko Ichimura, and Petra E. Todd, 1997, Matching as an econometric
evaluation estimator: Evidence from evaluating a job training programme, Review of Economic
Studies 64, 605–654.
Heckman, James J., Hidehiko Ichimura, and Petra E. Todd, 1998b, Matching as an econometric
evaluation estimator, Review of Economic Studies 65, 261–294.
Heckman, James J., Robert J. LaLonde, and Jeffrey Smith, 2000, The economics and econometrics
of active labor markets programs, in Orley C. Ashenfelter and David Card, (eds.) Handbook of
Labor Economics, volume 3, chapter 31 (Elsevier Science).
Heckman, James J., and R. Robb, 1986, Alternative methods for solving the problem of selection
bias in evaluating the impact of treatments on outcomes, in Howard Wainer, (ed.) Drawing
Inferences from Self-selected Samples, 63–107 (Springer-Verlag).
LoPucki, Lynn M., 2005, Protocols for the bankruptcy research database, Unpublished Manuscript,
June 2005 Version, UCLA School of Law.
LoPucki, Lynn M., and Joseph W. Doherty, 2004, The determinants of professional fees in large
bankruptcy reorganization cases, Journal of Empirical Legal Studies 1, 111–141.
Maksimovic, Vojislav, and Gordon Phillips, 1998, Asset efficiency and reallocation decisions of
bankrupt firms, Journal of Finance 53, 1495–1532.
Myers, Stewart C., and Nicholas Majluf, 1984, Corporate financing and investment decisions when
firms have information that investors do not have, Journal of Finance 39, 575–592.
Resnick, Alan N., and Henry J. Sommer, (eds.) , 2006, Collier Pamphlet Edition, Part 1, Bankruptcy
Code (LexisNexis).
Rosenbaum, Paul R., and Donald B. Rubin, 1983, The central role of the propensity score in
observational studies for causal effects, Biometrika 70, 41–55.
Smith, Jeffrey, and Petra Todd, 2005, Does matching overcome lalonde’s critique of nonexperimental
estimators?, Journal of Econometrics 125, 305–353.
32
White, Michelle J., 1989, The corporate bankruptcy decision, Journal of Economic Perspectives 3,
129–151.
Wruck, Karen Hopper, 1990, Financial distress, reorganization, and organizational efficiency, Jour-
nal of Financial Economics 27, 419–444.
Yost, Keven, 2002, The choice among traditional chapter 11, prepackaged bankruptcy, and out-of-
court restructuring, Working Paper, University of Wisconsin - Madison.
33
Tab
leI:
vari
able
sse
lect
edto
captu
refirm
s’ch
arac
teri
stic
saff
ecti
ng
ban
kru
ptc
yfiling
dec
isio
ns
chara
cter
isti
cspro
xy
vari
able
sit
em#
des
crip
tion
pro
fita
bility
sale
s/to
talass
ets
12/6
net
sale
sdiv
ided
by
tota
lass
ets
cash
flow
s/to
talass
ets
(14+
18)/
6aft
er-t
ax
cash
flow
sdiv
ided
by
tota
lass
ets
liquid
ity
quic
kra
tio
(1+
2)/
5quic
kass
ets
(cash
and
rece
ivable
s)div
ided
by
curr
ent
liability
EB
IT/cu
rren
tin
tere
stex
pen
ses
178/15
EB
ITst
ands
for
Earn
ings
Bef
ore
Inte
rest
and
Tax
outs
ide
financi
ng
fixed
ass
ets/
tota
lliability
8/181
fixed
ass
ets
are
tota
lpro
per
ty,pla
nt
and
equip
men
t
equity
valu
em
ark
etva
lue
ofeq
uity/to
talliability
(25*199)/
181
mark
etva
lue
ofco
mm
on
equity
norm
alize
dby
tota
lliability
liquid
ati
on
valu
ein
tangib
leass
ets/
tota
lass
ets
33/6
inta
ngib
les
are
ass
ets
that
hav
eno
physi
calex
iste
nce
inth
emse
lves
,but
repre
sent
rights
toen
joy
som
epri
vileg
e
barg
ain
ing
pro
ble
mse
cure
ddeb
tra
tio
241/181
secu
red
deb
tdiv
ided
by
tota
lliability
trade
cred
itra
tio
70/181
trade
cred
itis
captu
red
by
the
acc
ounts
pay
able
info
rmati
on
pro
ble
maudit
or’
sopin
ion
149
audit
ing
firm
’sopin
ion
regard
ing
aco
mpany’s
financi
alst
ate
men
ts
34
Table II: Descriptive Statistics for Bankruptcy Research Database
year number of cases for different categories months in bankruptcy for emergedtotal emerge non-emerge pending N.A. median mean minimum maximum
1980 3 3 0 0 0 38 33 23 381981 5 4 1 0 0 43.5 46.5 26 731982 13 11 2 0 0 35 37.5 15 831983 7 5 1 0 1 28 25.6 11 361984 6 6 0 0 0 28.5 27 14 361985 7 6 0 0 1 35.5 35.8 16 681986 10 8 2 0 0 25 30 2 821987 7 6 1 0 0 17.5 20.7 11 401988 12 11 1 0 0 24 24.4 13 441989 16 9 6 0 1 17 31.9 4 701990 30 21 9 0 0 23 23.2 4 381991 36 29 7 0 0 18 22.7 1 801992 32 26 6 0 0 8 10 1 481993 23 19 4 0 0 6 10.2 1 481994 11 7 4 0 0 1 7.3 1 201995 17 14 3 0 0 15 16.2 1 541996 13 5 8 0 0 2 4.6 1 141997 16 10 6 0 0 6 11.4 1 391998 30 18 10 1 1 11 12.6 2 251999 39 21 17 0 1 8 12.4 1 322000 69 30 35 4 0 14 13.6 1 312001 89 35 46 8 0 10 11.6 2 322002 75 42 23 7 3 6 8.6 1 242003 53 23 16 14 0 7 8.2 1 24all yrs 619 369 208 34 8 13 16.2 1 83
breakdown of the 208 non-emerged casestotal §363 sales converted to Ch. 7 liquidated or acquired dismissed
208 62 24 120 2
35
Table III: summary statistics of pre-filing characteristics
This table reports summary statistics of pre-filing characteristics for treatment and comparison
firm-years. The treatment group consists of large public firms from Bankruptcy Research Database
that have assets value over $100 million at the time of filing, measured in 1980 dollars, filed
Chapter 11 bankruptcy between 1980 and 2003, and eventually emerged. The comparison units
are extracted from COMPUSTAT that (1) are not identified in Bankruptcy Research Database,
(2) do not become inactive in COMPUSTAT due to Chapter 11 or Chapter 7, and (3) are large
corporations with assets worth more than $100 million, measured in 1980 dollars, for most of the
years they appear in the sample period. The empirical work is also done after eliminating outliers.
To do that, the observations in the comparison units that are larger and less than 99th and 1st
percentiles, respectively, are deleted for the following two variables: EBIT/total assets, market
value of equity/total liability. The empirical results for matching methods below are qualitatively
the same.
groups N mean median s.d. min max
sales/total assetstreatment 143 1.2121 1.0683 0.9394 0.0027 7.8382comparison 18397 1.1525 1.0086 0.8129 0.0000 9.6193
cash flows/total assetstreatment 143 -0.1204 -0.0180 0.2905 -1.5220 0.3542comparison 18397 0.0899 0.0931 0.0969 -3.2653 2.4222
quick ratiotreatment 143 0.6980 0.5537 0.6247 0.0290 3.3521comparison 18397 1.1723 0.9375 1.7053 0.0007 136.9843
EBIT/current interest expensestreatment 143 -0.1699 0.1571 2.7396 -22.8325 5.4548comparison 18397 25.0299 3.7969 726.4677 -1494.4 72751.8
fixed assets/total liabilitytreatment 143 0.4658 0.4203 0.3019 0.0087 1.3604comparison 18397 0.7806 0.6596 0.7740 0 46.3039
market value of equity/total liabilitytreatment 143 0.2497 0.1057 0.4043 0.0000 2.0204comparison 18397 2.9737 1.2780 54.4229 0.0000 7154.365
intangible assets/total assetstreatment 143 0.1371 0.0632 0.1687 0 0.8044comparison 18397 0.0947 0.0274 0.1482 0 0.9461
secured debt ratiotreatment 143 0.2008 0.1157 0.2206 0 0.8996comparison 18397 0.1022 0.0169 0.1742 0 0.9833
trade credit ratiotreatment 143 0.1115 0.0792 0.1019 0 0.5472comparison 18397 0.1669 0.1357 0.1336 0 1
auditor’s opiniontreatment 143 3.15 3 0.92 1 4comparison 18397 3.75 4 0.50 1 4
36
Tab
leIV
:es
tim
atio
nre
sult
sfo
rth
etr
adit
ional
mat
chin
gm
ethod
Each
row
offer
sth
ere
sult
sfo
rK
-nea
rest
-nei
ghborm
atc
hin
ges
tim
ate
s,fo
rK
=1,4
,6,1
0.
Panel
Aco
nsi
der
stw
oty
pes
ofoutc
om
em
easu
res:
sale
s
over
tota
lass
ets,
cash
flow
sov
erto
talass
ets,
all
mea
sure
din
the
post
-em
ergen
ceyea
r.Panel
Bm
easu
res
the
“outc
om
e”as
the
diff
eren
ced
valu
eofsa
les/
tota
lass
ets
and
cash
flow
s/to
talass
ets.
Inpart
icula
r,th
ediff
eren
ceis
calc
ula
ted
as
the
valu
ein
the
post
-em
ergen
ceyea
rm
inus
thatin
the
pre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
ris
the
firs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
6co
lum
ns.
The
firs
tco
lum
n
pre
sents
the
num
ber
ofm
atc
hed
nei
ghbors
.T
he
seco
nd
colu
mn
pro
vid
esth
enum
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)fo
rm
atc
hin
g.
The
thir
d
colu
mn
conta
ins
the
mea
noutc
om
esobse
rved
for
all
trea
tmen
tunit
s;th
efo
urt
hth
ees
tim
ate
dm
ean
counte
rfact
ualoutc
om
es,w
hic
hare
the
mea
noutc
om
esfo
rK
matc
hed
com
pari
son
unit
s.C
olu
mn
five
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
h
isco
lum
nth
ree
min
us
colu
mn
four.
The
last
colu
mn
conta
ins
the
boots
trapped
standard
erro
rsfo
rA
TT
esti
mato
rsbase
don
1000
replica
tions.
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
als
oobta
ined
for
1000
replica
tions.∗
den
ote
s
that
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
sre
ject
ion
at
5%
level
,∗∗∗
at
1%
level
.
PA
NEL
A
sale
s/to
talas
sets
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
114
31.
447
1.15
60.
291∗∗∗
(0.0
76)
114
30.
048
0.06
4-0
.016
(0.0
91)
414
31.
447
1.16
10.
286∗∗∗
(0.0
72)
414
30.
048
0.06
2-0
.014
(0.0
95)
614
31.
447
1.16
50.
282∗∗∗
(0.0
69)
614
30.
048
0.06
0-0
.012
(0.0
95)
1014
31.
447
1.14
90.
298∗∗∗
(0.0
69)
1014
30.
048
0.06
1-0
.013
(0.0
95)
PA
NEL
B
diffe
renc
eof
sale
s/to
talas
sets
diffe
renc
eof
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
114
30.
235
0.07
10.
164∗∗∗
(0.0
68)
114
30.
168
0.07
20.
096
(0.0
93)
414
30.
235
0.06
10.
174∗∗∗
(0.0
63)
414
30.
168
0.05
10.
117∗
(0.0
95)
614
30.
235
0.04
80.
187∗∗∗
(0.0
63)
614
30.
168
0.04
20.
126∗
(0.0
96)
1014
30.
235
0.04
10.
194∗∗∗
(0.0
63)
1014
30.
168
0.03
30.
135∗∗
(0.0
96)
37
Table V: Logistic regression estimationExplanatory variables are the pre-filing characteristics. Detailed explanation for each variableis documented in Table II. The dummy variables dummy2 to dummy4 correspond to scores2-4 assigned to the auditor’s opinion variable, respectively. The independent variable is equalto 1 if that pre-filing observation is a filing firm, equal to 0 otherwise.
characteristics proxy variables coefficients s.d. p-valueprofitability sales/total assets 0.1957 0.1043 0.0610
cash flows/total assets -1.5685 0.3412 0.0000
liquidity quick ratio -0.2692 0.1912 0.1590EBIT/current interest expenses -0.0107 0.0046 0.0200
outside financing fixed assets/total liability -0.3104 0.3447 0.3680
equity value market value of equity/total liability -3.8325 0.4031 0.0000
liquidation value intangible assets/total assets 3.7413 1.5207 0.0140square of intangible/total assets -4.6803 2.5762 0.0690
bargaining problem secured debt ratio 0.3193 0.4238 0.4510trade credit ratio -1.5658 1.0209 0.1250
information problem auditor’s opinion dummy2 0.6893 0.8532 0.4190dummy3 -0.9919 0.3261 0.0020dummy4 -1.7227 0.3291 0.0000
constant -1.0430 0.4255 0.0140
38
Tab
leV
I:es
tim
atio
nre
sult
sfo
rth
em
atch
ing
met
hod
bas
edon
the
pro
pen
sity
scor
eIn
each
panel
,th
efirs
t4
row
spre
sent
resu
lts
for
K-n
eare
st-n
eighbor
matc
hin
ges
tim
ati
on,w
her
eK
=1,4
,6,1
0.
The
last
row
conta
ins
resu
lts
for
ker
nel
-base
dm
atc
hin
g.
The
ker
nel
esti
mati
on
isim
ple
men
ted
wit
ha
bandw
idth
para
met
erof0.5
and
anorm
alker
nel
funct
ion.
Panel
Aco
nsi
der
stw
oty
pes
ofoutc
om
em
easu
res:
sale
sov
erto
talass
ets,
cash
flow
sov
erto
talass
ets,
all
mea
sure
din
the
post
-em
ergen
ceyea
r.Panel
Bm
easu
res
the
“outc
om
e”as
the
diff
eren
ced
valu
eofsa
les/
tota
lass
ets
and
cash
flow
s/to
talass
ets.
Inpart
icula
r,th
ediff
eren
ceis
calc
ula
ted
as
the
valu
ein
the
post
-em
ergen
ceyea
rm
inusth
atin
the
pre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
ris
the
firs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
6co
lum
ns.
The
firs
tco
lum
npre
sents
the
num
ber
ofm
atc
hed
nei
ghbors
.T
he
seco
nd
colu
mn
pro
vid
esth
enum
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)fo
rm
atc
hin
g.
The
thir
dco
lum
nco
nta
ins
the
mea
noutc
om
esobse
rved
for
all
trea
tmen
tunit
s;th
efo
urt
hth
ees
tim
ate
dm
ean
counte
rfact
ualoutc
om
es,w
hic
hare
the
mea
noutc
om
esfo
rK
matc
hed
com
pari
son
unit
s.C
olu
mn
five
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
his
colu
mn
thre
em
inus
colu
mn
four.
The
last
colu
mn
conta
ins
the
boots
trapped
standard
erro
rsfo
rA
TT
esti
mato
rsbase
don
1000
replica
tions.
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
als
oobta
ined
for
1000
replica
tions.
∗den
ote
sth
at
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
sre
ject
ion
at
5%
level
,∗∗∗
at
1%
level
.
PA
NEL
A
sale
s/to
talas
sets
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
114
31.
447
1.18
30.
264∗∗∗
(0.1
12)
114
30.
048
0.04
9-0
.001
(0.0
96)
414
31.
447
1.33
10.
116
(0.0
91)
414
30.
048
0.05
9-0
.011
(0.0
96)
614
31.
447
1.31
40.
133∗
(0.0
88)
614
30.
048
0.05
7-0
.009
(0.0
96)
1014
31.
447
1.27
70.
170∗∗
(0.0
84)
1014
30.
048
0.05
7-0
.009
(0.0
96)
ker
nel
143
1.44
71.
103
0.34
4∗∗∗
(0.0
87)
ker
nel
143
0.04
80.
089
-0.0
41(0
.096
)
PA
NEL
B
diffe
renc
eof
sale
s/to
talas
sets
diffe
renc
eof
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
114
30.
235
0.02
80.
207∗∗∗
(0.0
70)
114
30.
168
0.11
40.
054
(0.0
98)
414
30.
235
0.07
90.
156∗∗∗
(0.0
62)
414
30.
168
0.09
00.
078
(0.0
98)
614
30.
235
0.08
00.
155∗∗∗
(0.0
62)
614
30.
168
0.08
90.
079
(0.0
98)
1014
30.
235
0.07
90.
156∗∗∗
(0.0
62)
1014
30.
168
0.07
10.
097
(0.0
98)
ker
nel
143
0.23
5-0
.034
0.26
9∗∗∗
(0.0
62)
ker
nel
143
0.16
8-0
.003
0.17
1∗∗
(0.0
98)
39
Tab
leV
II:es
tim
atio
nre
sult
sfo
rth
em
atch
ing
met
hod
bas
edon
the
pro
pen
sity
scor
e(w
ith
com
mon
suppor
t)In
each
panel
,th
efirs
t4
row
spre
sent
resu
lts
for
K-n
eare
st-n
eighbor
matc
hin
ges
tim
ati
on,w
her
eK
=1,4
,6,1
0.
The
last
row
conta
ins
resu
lts
for
ker
nel
-base
dm
atc
hin
g.
The
ker
nel
esti
mati
on
isim
ple
men
ted
wit
ha
bandw
idth
para
met
erof0.5
and
anorm
alker
nel
funct
ion.
Com
mon
support
isim
pose
dusi
ng
the
alg
ori
thm
outl
ined
inth
eA
ppen
dix
B.
The
trim
min
gru
leq
isse
tto
be
0.0
1.
Panel
Aco
nsi
der
stw
oty
pes
of
outc
om
em
easu
res:
sale
sov
erto
tal
ass
ets,
cash
flow
sov
erto
tal
ass
ets,
all
mea
sure
din
the
post
-em
ergen
ceyea
r.Panel
Bm
easu
res
the
“outc
om
e”as
the
diff
eren
ced
valu
eof
sale
s/to
tal
ass
ets
and
cash
flow
s/to
tal
ass
ets.
Inpart
icula
r,th
ediff
eren
ceis
calc
ula
ted
as
the
valu
ein
the
post
-em
ergen
ceyea
rm
inus
that
inth
epre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
ris
the
firs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
6co
lum
ns.
The
firs
tco
lum
npre
sents
the
num
ber
ofm
atc
hed
nei
ghbors
.T
he
seco
nd
colu
mn
pro
vid
esth
enum
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)fo
rm
atc
hin
g.
The
thir
dco
lum
nco
nta
ins
the
mea
noutc
om
esobse
rved
for
all
trea
tmen
tunit
s;th
efo
urt
hth
ees
tim
ate
dm
ean
counte
rfact
ualoutc
om
es,w
hic
hare
the
mea
noutc
om
esfo
rK
matc
hed
com
pari
son
unit
s.C
olu
mn
five
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
his
colu
mn
thre
em
inus
colu
mn
four.
The
last
colu
mn
conta
ins
the
boots
trapped
standard
erro
rsfo
rA
TT
esti
mato
rsbase
don
1000
replica
tions.
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
als
oobta
ined
for
1000
replica
tions.
∗den
ote
sth
at
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
sre
ject
ion
at
5%
level
,∗∗∗
at
1%
level
.
PA
NEL
A
sale
s/to
talas
sets
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
110
11.
383
1.14
20.
241∗∗
(0.1
39)
110
10.
023
0.04
1-0
.018
(0.1
30)
410
11.
383
1.34
50.
038
(0.1
20)
410
10.
023
0.05
9-0
.036
(0.1
30)
610
11.
383
1.34
30.
040
(0.1
09)
610
10.
023
0.06
0-0
.037
(0.1
30)
1010
11.
383
1.28
30.
100
(0.1
04)
1010
10.
023
0.06
0-0
.037
(0.1
30)
ker
nel
101
1.38
31.
103
0.28
0∗∗∗
(0.1
04)
ker
nel
101
0.02
30.
088
-0.0
65(0
.130
)
PA
NEL
B
diffe
renc
eof
sale
s/to
talas
sets
diffe
renc
eof
cash
flow
s/to
talas
sets
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
nei
ghbor
#of
obse
rved
counte
r.es
tim
ate
ds.
e.of
obs.
mea
nm
ean
AT
TA
TT
obs.
mea
nm
ean
AT
TA
TT
110
10.
236
-0.0
160.
252∗∗∗
(0.0
90)
110
10.
054
0.03
90.
015
(0.1
31)
410
10.
236
0.03
70.
199∗∗∗
(0.0
81)
410
10.
054
0.04
60.
008
(0.1
30)
610
10.
236
0.05
30.
183∗∗
(0.0
79)
610
10.
054
0.04
80.
006
(0.1
31)
1010
10.
236
0.05
40.
182∗∗
(0.0
80)
1010
10.
054
0.03
90.
015
(0.1
31)
ker
nel
101
0.23
6-0
.043
0.27
9∗∗∗
(0.0
79)
ker
nel
101
0.05
4-0
.008
0.06
2(0
.131
)
40
Tab
leV
III:
esti
mat
ion
resu
lts
for
the
sub-s
ample
ofpos
t-19
90filing
case
s
This
table
pro
vid
eses
tim
ati
on
resu
lts
for
the
sub-s
am
ple
ofpost
-1990
filing
case
s.In
each
panel
,th
efirs
t4
row
spre
sent
resu
lts
for
K-n
eare
st-
nei
ghbor
matc
hin
ges
tim
ati
on,w
her
eK
=1,4
,6,1
0.
The
last
row
inPanel
Bco
nta
ins
resu
lts
for
ker
nel
-base
dm
atc
hin
g.
The
ker
nel
esti
mati
on
isim
ple
men
ted
wit
ha
bandw
idth
para
met
erof0.5
and
anorm
alker
nel
funct
ion.
“N
”re
pre
sents
the
num
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)
for
matc
hin
g.
Inea
chpanel
,th
efirs
ttw
om
ajo
rco
lum
ns
pre
sent
resu
lts
for
outc
om
eva
riable
sm
easu
red
inth
epost
-em
ergen
ceyea
r.T
he
last
two
majo
rco
lum
ns
mea
sure
the
outc
om
eva
riable
as
the
diff
eren
ced
valu
ebet
wee
nth
epost
-em
ergen
ceyea
rand
the
pre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
ris
the
firs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
3co
lum
ns.
Colu
mn
(1)
conta
ins
the
mea
noutc
om
es
obse
rved
for
all
trea
tmen
tunit
s;co
lum
n(2
)pre
sents
the
esti
mate
dm
ean
counte
rfact
ual
outc
om
es,
whic
hare
the
mea
noutc
om
esfo
rall
matc
hed
com
pari
son
unit
s.C
olu
mn
(3)
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
his
(1)
min
us
(2).
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
obta
ined
for
1000
replica
tions.
∗den
ote
sth
at
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
sre
ject
ion
at
5%
level
,∗∗∗
at
1%
level
. outc
om
em
easu
rein
the
post
-em
ergen
ceyea
rdiff
eren
cebet
wee
nth
epre
-filing
and
post
-em
ergen
ceyea
rs
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
neig
hborN
obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.
mean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
T
PA
NEL
A.Tradit
ionalm
atc
hin
gest
imate
s
111
41.
414
1.08
20.
332∗∗∗
0.05
10.
062
-0.0
110.
235
0.06
20.
173∗∗
0.19
00.
075
0.11
54
114
1.41
41.
129
0.28
5∗∗∗
0.05
10.
062
-0.0
110.
235
0.06
70.
168∗∗∗
0.19
00.
045
0.14
5∗
611
41.
414
1.12
10.
293∗∗∗
0.05
10.
057
-0.0
060.
235
0.06
70.
168∗∗∗
0.19
00.
037
0.15
3∗
1011
41.
414
1.10
60.
308∗∗∗
0.05
10.
057
-0.0
060.
235
0.05
80.
177∗∗∗
0.19
00.
032
0.15
8∗
PA
NEL
B.P
ropensi
tysc
ore
matc
hin
gest
imate
s
111
41.
414
1.26
90.
145∗
0.05
10.
058
-0.0
070.
235
0.09
50.
140∗∗
0.19
00.
132
0.05
84
114
1.41
41.
338
0.07
60.
051
0.05
8-0
.007
0.23
50.
100
0.13
5∗∗
0.19
00.
095
0.09
56
114
1.41
41.
276
0.13
8∗∗
0.05
10.
056
-0.0
050.
235
0.08
60.
149∗∗
0.19
00.
088
0.10
210
114
1.41
41.
270
0.14
4∗∗
0.05
10.
057
-0.0
060.
235
0.08
80.
147∗∗
0.19
00.
074
0.11
6ker
nel
114
1.41
41.
076
0.33
8∗∗∗
0.05
10.
087
-0.0
360.
235
-0.0
240.
259∗∗∗
0.19
0-0
.003
0.19
3
41
Tab
leIX
:es
tim
atio
nre
sult
sfo
rth
esu
b-s
ample
offiling
case
sth
atst
ayin
ban
kru
ptc
yfo
rle
ssth
an1
year
This
table
pro
vid
eses
tim
ati
on
resu
lts
for
the
sub-s
am
ple
offiling
case
sth
at
stay
inbankru
ptc
yfo
rle
ssth
an
1yea
r.In
each
panel
,th
efirs
t4
row
spre
sent
resu
lts
for
K-n
eare
st-n
eighbor
matc
hin
ges
tim
ati
on,w
her
eK
=1,4
,6,1
0.
The
last
row
inPanel
Bco
nta
ins
resu
lts
for
ker
nel
-base
d
matc
hin
g.
The
ker
nel
esti
mati
on
isim
ple
men
ted
wit
ha
bandw
idth
para
met
erof0.5
and
anorm
alker
nel
funct
ion.
“N
”re
pre
sents
the
num
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)fo
rm
atc
hin
g.
Inea
chpanel
,th
efirs
ttw
om
ajo
rco
lum
ns
pre
sent
resu
lts
for
outc
om
eva
riable
sm
easu
red
inth
e
post
-em
ergen
ceyea
r.T
he
last
two
majo
rco
lum
ns
mea
sure
the
outc
om
eva
riable
as
the
diff
eren
ced
valu
ebet
wee
nth
epost
-em
ergen
ceyea
rand
the
pre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
r
isth
efirs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
3co
lum
ns.
Colu
mn
(1)
conta
ins
the
mea
noutc
om
esobse
rved
for
all
trea
tmen
tunit
s;co
lum
n(2
)pre
sents
the
esti
mate
dm
ean
counte
rfact
ualoutc
om
es,w
hic
hare
the
mea
n
outc
om
esfo
rall
matc
hed
com
pari
son
unit
s.C
olu
mn
(3)
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
his
(1)
min
us
(2).
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
obta
ined
for
1000
replica
tions.
∗
den
ote
sth
at
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
s
reje
ctio
nat
5%
level
,∗∗∗
at
1%
level
.
outc
om
em
easu
rein
the
post
-em
ergen
ceyea
rdiff
eren
cebet
wee
nth
epre
-filing
and
post
-em
ergen
ceyea
rs
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
neig
hborN
obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.
mean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
T
PA
NEL
A.Tradit
ionalm
atc
hin
gest
imate
s
171
1.42
01.
228
0.19
2∗0.
079
0.05
90.
020
0.24
40.
051
0.19
3∗∗
0.24
70.
102
0.14
54
711.
420
1.20
10.
219∗∗∗
0.07
90.
053
0.02
60.
244
0.07
20.
172∗∗
0.24
70.
073
0.17
46
711.
420
1.17
90.
241∗∗∗
0.07
90.
055
0.02
40.
244
0.07
10.
173∗∗
0.24
70.
064
0.18
310
711.
420
1.14
30.
277∗∗∗
0.07
90.
059
0.02
00.
244
0.05
90.
185∗∗
0.24
70.
058
0.18
9
PA
NEL
B.P
ropensi
tysc
ore
matc
hin
gest
imate
s
171
1.42
01.
190
0.23
0∗0.
079
0.05
80.
021
0.24
40.
044
0.20
0∗∗
0.24
70.
178
0.06
94
711.
420
1.24
90.
171∗
0.07
90.
049
0.03
00.
244
0.09
00.
154∗∗
0.24
70.
126
0.12
16
711.
420
1.21
10.
209∗∗
0.07
90.
049
0.03
00.
244
0.11
20.
132∗
0.24
70.
130
0.11
710
711.
420
1.19
10.
229∗∗
0.07
90.
054
0.02
50.
244
0.08
10.
163∗∗
0.24
70.
098
0.14
9ker
nel
711.
420
1.06
30.
357∗∗∗
0.07
90.
085
-0.0
060.
244
-0.0
150.
259∗∗∗
0.24
7-0
.001
0.24
8∗
42
Tab
leX
:es
tim
atio
nre
sult
sfo
rth
esu
b-s
ample
offiling
case
sth
atst
ayin
ban
kru
ptc
yfo
rm
ore
than
1ye
ar
This
table
pro
vid
eses
tim
ati
on
resu
lts
for
the
sub-s
am
ple
offiling
case
sth
at
stay
inbankru
ptc
yfo
rm
ore
than
1yea
r.In
each
panel
,th
efirs
t4
row
spre
sent
resu
lts
for
K-n
eare
st-n
eighbor
matc
hin
ges
tim
ati
on,w
her
eK
=1,4
,6,1
0.
The
last
row
inPanel
Bco
nta
ins
resu
lts
for
ker
nel
-base
d
matc
hin
g.
The
ker
nel
esti
mati
on
isim
ple
men
ted
wit
ha
bandw
idth
para
met
erof0.5
and
anorm
alker
nel
funct
ion.
“N
”re
pre
sents
the
num
ber
oftr
eatm
ent
unit
s(fi
ling
firm
s)fo
rm
atc
hin
g.
Inea
chpanel
,th
efirs
ttw
om
ajo
rco
lum
ns
pre
sent
resu
lts
for
outc
om
eva
riable
sm
easu
red
inth
e
post
-em
ergen
ceyea
r.T
he
last
two
majo
rco
lum
ns
mea
sure
the
outc
om
eva
riable
as
the
diff
eren
ced
valu
ebet
wee
nth
epost
-em
ergen
ceyea
rand
the
pre
-filing
yea
r.T
he
post
-em
ergen
ceyea
ris
the
firs
tyea
raft
erth
efiling
firm
s’em
ergen
cefo
rw
hic
hdata
are
available
.T
he
pre
-filing
yea
r
isth
efirs
tyea
rbef
ore
the
firm
s’filing
for
whic
hdata
are
available
.P
rovid
edfo
rea
choutc
om
em
easu
reare
3co
lum
ns.
Colu
mn
(1)
conta
ins
the
mea
noutc
om
esobse
rved
for
all
trea
tmen
tunit
s;co
lum
n(2
)pre
sents
the
esti
mate
dm
ean
counte
rfact
ualoutc
om
es,w
hic
hare
the
mea
n
outc
om
esfo
rall
matc
hed
com
pari
son
unit
s.C
olu
mn
(3)
pro
vid
esth
ees
tim
ate
dA
ver
age
Tre
atm
ent
effec
ton
the
Tre
ate
d(A
TT
),w
hic
his
(1)
min
us
(2).
To
ass
ess
the
stati
stic
alsi
gnifi
cance
,boots
trap
dis
trib
uti
on
and
reje
ctio
nre
gio
ns
oft-
stati
stic
are
obta
ined
for
1000
replica
tions.
∗
den
ote
sth
at
we
can
reje
ct,at
10%
signifi
cance
level
,th
enull
hypoth
esis
infa
vor
ofth
ealt
ernati
ve
hypoth
esis
that
AT
Tis
posi
tive.∗∗
den
ote
s
reje
ctio
nat
5%
level
,∗∗∗
at
1%
level
.
outc
om
em
easu
rein
the
post
-em
ergen
ceyea
rdiff
eren
cebet
wee
nth
epre
-filing
and
post
-em
ergen
ceyea
rs
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
sale
s/to
talass
ets
cash
flow
s/to
talass
ets
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
neig
hborN
obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.obs.
counte
r.est
.
mean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
Tm
ean
mean
AT
T
PA
NEL
A.Tradit
ionalm
atc
hin
gest
imate
s
172
1.47
31.
145
0.32
8∗∗∗
0.01
80.
062
-0.0
440.
226
0.08
60.
140
0.09
10.
024
0.06
7∗∗∗
472
1.47
31.
154
0.31
9∗∗∗
0.01
80.
067
-0.0
490.
226
0.08
40.
142∗
0.09
10.
023
0.06
8∗∗∗
672
1.47
31.
140
0.33
3∗∗∗
0.01
80.
070
-0.0
520.
226
0.04
50.
181∗∗
0.09
10.
022
0.06
9∗∗∗
1072
1.47
31.
132
0.34
1∗∗∗
0.01
80.
069
-0.0
510.
226
0.03
30.
193∗∗
0.09
10.
018
0.07
3∗∗∗
PA
NEL
B.P
ropensi
tysc
ore
matc
hin
gest
imate
s
172
1.47
31.
184
0.28
9∗∗
0.01
80.
046
-0.0
280.
226
0.02
90.
197∗∗
0.09
10.
019
0.07
2∗∗
472
1.47
31.
301
0.17
2∗0.
018
0.06
1-0
.043
0.22
60.
023
0.20
3∗∗
0.09
10.
028
0.06
3∗∗
672
1.47
31.
363
0.11
00.
018
0.05
8-0
.040
0.22
60.
034
0.19
2∗∗
0.09
10.
027
0.06
4∗∗
1072
1.47
31.
410
0.06
30.
018
0.06
1-0
.043
0.22
60.
054
0.17
2∗∗
0.09
10.
029
0.06
2∗∗
ker
nel
721.
473
1.13
90.
334∗∗∗
0.01
80.
093
-0.0
750.
226
-0.0
550.
281∗∗∗
0.09
1-0
.007
0.09
8∗∗∗
43