Journal of Business Finance & Accounting, 29(3) & (4), April/May 2002, 0306-686X

Predicting Bankruptcy Resolution

Ran Barniv, Anurag Agarwal and Robert Leach*

1. INTRODUCTION AND BACKGROUND

Following bankruptcy filing, financially distressed firms in the

U.S. are placed under court supervision, until the bankruptcy is
resolved. Filing for corporate bankruptcy is required under
Chapter 11 of the 1978 bankruptcy code, where management
and owners seek court protection against creditors and other
claimants.1 Usually the court confirms a reorganization or
rehabilitation plan following the bankruptcy filing. With the
approval of the court, bankruptcy filing of publicly traded firms is
finally settled, on the average about 17 months after the filing, in
the following three alternative resolutions. Firms are either (i)
acquired by other firms or (ii) emerged as independent entities
or (iii) liquidated.
Firms filing for bankruptcy share similar characteristics of
financial distress and therefore it is difficult to predict the final
resolution. For example, prior to filing bankruptcy, most firms
disclose declining revenues, total assets, book value of equities
and earnings. Thus, we expect that distinguishing between the
three post-bankruptcy filing outcomes is more difficult than
discriminating between sound and financially distressed firms,
which has been extensively examined in prior literature.
The purpose of this paper is to classify and predict the final
bankruptcy resolution. This research issue has not been
examined in prior accounting and finance literature. We develop

* The authors are respectively from Kent State University, the University of Florida and the
University of South Carolina at Aiken. They acknowledge the helpful comments of an
anonymous referee and an Executive Board referee. Robert Leach provided extensive
data collection efforts. (Paper received January 2000, revised and accepted February 2001)
Address for correspondence: Ran Barniv, Graduate School of Management, College of
Business Administration, Kent State University, Kent, Ohio, USA.
e-mail: [email protected]

ß Blackwell Publishers Ltd. 2002, 108 Cowley Road, Oxford OX4 1JF, UK
and 350 Main Street, Malden, MA 02148, USA. 497
498 BARNIV, AGARWAL AND LEACH

a classification and prediction model using logistic regression to

classify and predict firms as acquired, emerged or liquidated (a
three-group resolution) and also as `acquired or emerged' versus
liquidated (a two-group resolution). A ten-variable logistic
regression model is developed employing financial accounting
data and non-financial data.2 The model includes five accounting
variables and five non-accounting variables used by bankruptcy
courts. Accounting variables include, among others, profitability
ratio and a proxy for size. Court related non-financial accounting
variables include resignation of top executives, fraud and other
characteristics. While the study focuses on classification and
prediction provided by the ten-variable model, some discussion
of the role of accounting and non-accounting variables is also
provided.
From the perspectives of investors, creditors and other
stakeholders it is important to predict the outcome following
bankruptcy filing. Our preliminary results suggest that investors
in filing firms lose significant market values from sixty days prior
to filing through one trading day after the filing. More
importantly, following the filing, investors in acquired and
emerged firms experience significant positive returns between
the filing and final resolution, whereas investors in liquidated
firms sustain significant additional negative returns. For example,
cumulative abnormal returns (CARs) for investors in acquired
and emerging firms are, on the average, 155 percent and 137
percent, respectively, from filing through the final court
resolution. CARs for liquidated firms, on the other hand, are
on average negative 11 percent. Therefore, the three resolutions
are ordered based on the returns to investors as acquired (best),
emerged (medium) and liquidated (worst). Since liquidation
generates negative CARs from filing through the final resolution
while acquisition and emergence generate positive CARs, we also
examine a two-group resolution of acquired or emerged vs.
liquidated firms. Other methods of calculating returns including
buy-and-hold returns provide the same ordering, though
different magnitudes of return.3 It is therefore economically
desirable to predict the post bankruptcy resolution event.
We use the entire list of firms filing for Chapter 11 between
1980 and 1995, provided by the Securities and Exchange
Commission (SEC) to generate our sample. From this list, 237

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 499

publicly traded firms, for which the final bankruptcy resolution is

already known, have complete financial accounting information,
non-financial court-related data and securities prices are selected.
For measuring classification and prediction accuracy we use rate
of correct classification and also briefly examine the expected
cost of misclassification (ECM).
The ten-variable logistic regression model performs quite well
in both classification and prediction tasks. For the entire sample,
the three-group classification accuracy is about 62 percent using
the Lachenbruch (1967) technique, while for a time-series
holdout sample, the classification accuracy is about 49 percent.
These classification accuracies are significantly better than that
expected from random classification.4
Some supplemental results suggest that a five-variable non-
accounting model performs slightly better than a five-variable
accounting model. Further, the classification accuracy and the
ECM of the ten-variable model are better than those of both the
five-variable models. These results indicate that financial
statement data alone is not adequate, and researchers as well as
external users of accounting data would benefit from using non-
financial accounting information for classifying and predicting
bankruptcy resolution.5
The remainder of this paper is organized as follows. In Section
2, we provide a brief literature review and discuss our incremen-
tal contribution. In Section 3 we present the research method-
ology which includes a discussion of ordered logistic regression
and classification accuracies. Data, sampling and variable
selection are described in a subsequent section, followed by a
section on empirical results. The summary and conclusions sec-
tion also includes a discussion of the public policy implications of
our results to investors, bankruptcy courts and other stake-
holders.

2. PRIOR LITERATURE AND INCREMENTAL CONTRIBUTION

Prior studies have focused on classifying and predicting financial

distress or the event of filing bankruptcy. Those studies focus on
two-group classification namely healthy versus financially
distressed firms. Numerous accounting and finance studies have

ß Blackwell Publishers Ltd 2002

500 BARNIV, AGARWAL AND LEACH

been published since the pioneering article by Altman (1968).

Subsequent early articles use primarily multiple discriminant
analysis (MDA) for predicting bankruptcy. Later studies use logit
(e.g., Ohlson, 1980), or probit (e.g., Zmijewski, 1984) for
predicting financial distress.6
Recent studies examine various issues related to predicting
financial distress and bankruptcy filing. For example, Richardson
et. al. (1998) examine the impact of recession on the prediction
of corporate failure. Ward and Foster (1997) use loan default/
accommodation as a response measure for financial distress.
Akhigbe and Madura (1996) examine the intra-industry effect on
voluntary corporate liquidations. Platt et. al. (1994) examine
bankruptcy discrimination with real variables. Hsieh (1993)
discusses optimal cutoff points in bankruptcy prediction models.
Other aspects of bankruptcy prediction and related issues are
also discussed (Ro et al., 1992; Tennyson et al., 1990; and Platt
and Platt, 1990). Measuring the classification accuracy has been
the primary issue in these and many other studies (e.g., Dopuch
et al., 1987; and Liang et al., 1992). Lau (1987) and Ward (1994)
develop, respectively, five-state and four-state models of financial
distress prior to filing. Recently Laitnen and Laitnen (1998) and
Dhumale (1998) use logistic regressions with cash management
model and earning retention, respectively, to predict corporate
failure. Lennox (1999) discusses the impact of the audit report
on bankruptcy prediction. None of these studies have examined
post-bankruptcy resolution.
Only a few studies directly explore aspects of the post-
bankruptcy scenario. LoPucki (1983) uses a small sample of
bankrupt firms to examine the outcome of bankruptcy
reorganization using only correlation analyses. Lehavy (1999)
examines reporting discretion and fresh start values for firms
emerging from bankruptcy. White (1983 and 1989) analyzes the
confirmation decision through modeling a coalition of creditors
and equity holders. Casey et. al. (1986) uses only accounting
variables to discriminate between a group of liquidated firms and
a group of restructured firms. Kim and Kim (1999) use a similar
two-group analysis applied to Korean firms.
Recent studies examine other aspects of post-bankruptcy filing
(e.g., Morse and Shaw, 1988; and Hotchkiss, 1995). However,
classification and prediction of the final outcome have not been

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 501

examined in this prior accounting and finance literature. Our

study differs from the earlier literature by focusing explicitly on
the classification and prediction of the post bankruptcy
resolution event using accounting and non-accounting variables
available at the beginning of the bankruptcy filing period. Such
predictions may be very relevant for investors in developing
investment strategies in bankrupt firms. Creditors and other
claimants would also benefit by being able to develop
appropriate strategies during the prolonged court deliberations.
Bankruptcy judges may also use these predictions as preliminary
indicators of the potential future outcome. These and other
public policy implications of our approach are discussed in the
summary and conclusion section.

3. RESEARCH METHODOLOGY

(i) Ordered Logistic Regression

Most recent studies use and compare multiple techniques for
classification such as logit, MDA and Neural Networks (NN). For
example, Ohlson (1980), Lo (1986) and Maddala (1991) provide
discussions on logit models and comparisons with other
techniques. For our study, we use logistic regression, which has
been the most commonly used technique in recent literature.7
We use ordered logistic regressions (Maddala, 1983 and 1991)
which fit the ranking of the final resolution (the dependent
variable) in ordered ranking of final resolution. As discussed in
the introduction, acquired is a better state than emerged, which
in turn, is better than a liquidated state. Therefore an ordered
logit, which has the following cumulative logistic function is used:
1
P …r jBF† ˆ F …r ‡ X 0 † ˆ ;
1‡ e ÿ…r ‡X 0 †
where r ˆ 1; 2 and 3 for the three group resolutions and 1and 2
for two-group resolutions; 3 > 2 > 1 ˆ 0; X 0  is the vector of
coefficients multiplied by the vector of variables, and P(r|BF) is
the conditional probability of the resolution, given that a firm
filed for bankruptcy.

ß Blackwell Publishers Ltd 2002

502 BARNIV, AGARWAL AND LEACH

(ii) Classification Accuracies

Different cutoff points have been used in prior studies for measur-
ing classification and prediction accuracy. For example, Altman
(1968) and Deakin (1972) use cutoff points which minimize
misclassification accuracy and Ohlson (1980) and Palepu (1986)
use the optimal cutoff points where the distributions of the two
groups intersect. For classification and prediction purposes we use
cutoff points that minimize the number of misclassifications. The
classification accuracy is examined on the full sample using the
Lachenbruch (1967) U-technique. This method holds out one
observation at a time, estimates the multivariate function with the
rest of the observations, and then classifies the holdout obser-
vation. This classification is repeated until all observations are
classified. In addition, classification accuracy is also examined on
an inter-temporal split basis using an estimation sample data
drawn from earlier years and a holdout sample with data from
subsequent years. The cutoff points, which minimize the number
of misclassifications for the estimation sample are used to
determine the classification accuracy in the holdout sample
(Dopuch et al., 1987).
Several studies use the cutoff points that minimize the ECM
(Frydman et al., 1985) because misclassifying firms of different
groups may have different costs. A general ECM for more than
two-state is presented as follows:
X
k
nr
ECM ˆ Pr Cqjr
r ˆ1
Nr
where, r is any type of resolution, q is a set of all resolutions not
equal to r, nr is the number of firms misclassified for rth
resolution. Nr is the sample size for the rth resolution. Pr is the
prior probability of resolution r and Cq|r is the cost of
misclassifying an observation belonging to group r into group q.
For the three-group resolution, a simplified equation is:
n1 n2 n3
ECM ˆ P1 C2;3j1 ‡ P2 C1;3j2 ‡ P3 C1;2j3 ;
N1 N2 N3
where ni is the total number of type i misclassifications; Ni is the
sample size of the ith state, i ˆ 1, 2 and 3 in this study; Pi equals the
prior probability for state i and C..|i equals the cost of

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 503

misclassification of an observation belonging to group i. This

equation can be further simplified for any two-group resolution.
Empirical analysis indicates that an investment of $100 in a filing
firm, one day after filing, returns over $200 for firms that were
eventually acquired or emerged and about $80 for liquidated
firms. Given these results, misclassifying a liquidated firm is
approximately three times more costly than misclassifying
acquired or emerged firms. Therefore we use C2;3j1 ˆ C1;3j2 ˆ 1
and C1;2j3 ˆ 3, which is also confirmed by the CARs reported in
the introduction.

4. DATA SAMPLING AND VARIABLES

(i) Sampling
We obtain the list of all publicly traded firms which filed
bankruptcy from 1980 through 1995 from the Office of the
General Counsel of the SEC. However, we use only those firms
that filed for bankruptcy prior to 1992 because some firms
remained under bankruptcy court supervision for over six years.
All companies which operate in regulated industries, such as
financial institutions, and some other types of firms are
excluded.8
A sample of 237 publicly traded firms with complete
accounting data, price data, and court-related data is used. Table
1 shows the distribution of the firms by industry. The majority of
the companies are manufacturing firms (99) and wholesale and
retail firms (61) which comprise 42 percent and 26 percent of the
sample, respectively. Data are obtained from COMPUSTAT and
supplemented from annual reports and 10-Ks available on
microfilm. Prices are obtained from the CRSP and are extended
using data provided by Wheat-First-Butcher-Singer for firms
traded in regional exchanges. Additional court related data are
obtained using Lexis/Nexis, Moody's Industrial Report, Wall Street
Journal and Commerce Clearing House's Capital Change Reporting.
The 237 filing firms include 49 (21 percent) acquired firms,
119 (50 percent) emerged firms, and 69 (29 percent) liquidated
firms. We analyze the final bankruptcy resolutions for the entire
population of all publicly traded firms that filed for bankruptcy

ß Blackwell Publishers Ltd 2002

504 BARNIV, AGARWAL AND LEACH

Table 1
The Sample of Bankrupt Firms

Industry Bankruptcy Resolution All

(two digit SIC code) Acquired Emerged Liquidated Firms

Agriculture, Mining 7 24 7 38 (16%)

and Construction
(01±19)
Manufacturing 20 46 33 99 (42%)
(20±39)
Wholesail and Retail 12 29 20 61 (26%)
(50±59)
Other Industries 10 20 9 39 (16%)
(40±46, 70±89)
Total Firms 49 (21%) 119 (50%) 69 (29%) 237 (100%)

between 1980 and 1992 obtained from the SEC. The proportions
of firms for the three resolutions in the entire population are
similar to the proportions in our sample. This full sample is used
to predict the final resolution at the first day after the filing. The
firms are further split into estimation and holdout samples. The
estimation sample includes 114 firms that filed for bankruptcy
from 1980 through 1986. The holdout sample includes 123 firms
that filed bankruptcy from 1987 through 1991.

(ii) Background
Before we discuss the variables it is important to understand the
sequence of events during the post-filing period because some of
the variables affect the specific events and are used by various
stakeholders throughout the bankruptcy process. For example,
holders of secured debt may oppose reorganization and promote
liquidation. A timeline of major events is depicted in Figure 1,
along with dates and CARs. The filing of Chapter 11 petition date
is usually followed by a filing of a schedule of assets and liabilities
with the court. This process is followed by bar deliberations
where creditors must file a proof of claim on the bar date. The
next important date is the filing of a reorganization plan, which is
followed by hearing and disclosure statements and a final voting
on the plan, confirmation hearing and final confirmation. If the

ß Blackwell Publishers Ltd 2002

ß Blackwell Publishers Ltd 2002

Figure 1
Timeline of Major Events, Dates and CARs

PREDICTING BANKRUPTCY RESOLUTION

Notes:
A: Annual Report and/or 10-K release
B: Filing Day
C: Filing Schedule of Assets and Liabilities
D: Filing a proof of claim at the bar date
E: Filing Reorganization Plan

505
F: Confirmation Date
*: The average duration between filing and resolution.
506 BARNIV, AGARWAL AND LEACH

reorganization plan is not confirmed, the bankruptcy judge can

rule to transfer the firm to liquidation under Chapter 7.
However, if organization continues, two major events may occur.
The firm may exit the bankruptcy process and emerge as a new
entity or it may be acquired by another firm.
We use the filing date and the final resolution date as the
major useful events. The duration between filing and final
resolution is, on the average 364 days for liquidated firms, 501
days for acquired firms, and 633 days for emerged firms.
Classification and prediction is done at the end of the first
trading day after the filing using financial and non-financial data
available to the court and to investors at that day. The CARs
obtained from filing through the final resolution are also
presented. The CARs for the acquired firms are higher than
those for emerged firms and the CARs for the liquidated firm are
negative and decrease between bankruptcy filing and the final
resolution event. These results further illustrate the differences
between the three groups. As explained before, Figure 1 provides
some background for the variable selection process described
below.

(iii) Variable Selection

To select financial accounting characteristics, we examine
variables used in three previous studies for predicting bankruptcy
(Altman, 1968; Ohlson, 1980; and Frydman et al., 1985).
Intuitively, these variables would be useful for predicting the
final resolution. Therefore our initial set includes these variables.
We apply the following variable selection process for these
variables. First, a sample of more than 3000 healthy firms and our
237 filing firms is used to examine the significance levels of the
estimated coefficients and the classification accuracies using the
obtained variables of each of these three models separately. The
results (not reported) demonstrate that all variables included in
each of these three studies perform reasonably well for
predicting the bankruptcy filing. However, in contrast, univariate
analysis of the variables and logistic regressions based on these
three studies suggest that all but two of these variables are not
statistically significant for predicting the final resolution. Only
the firm size (LNTA) and Net Income to Total Assets (NI/TA),

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 507

are found to be somewhat significant for various three groups or

two sub-group classifications. Therefore, in search for additional
relevant variables, we scrutinize more than twenty five accounting
and non-accounting court-related characteristics that have been
used in practice by bankruptcy judges and attorneys.9
A three step analysis is used for selecting the additional variables.
First, we conduct correlation analysis and for variables that are
highly correlated with one another, only one variable is selected.
Second, we use univariate analysis to exclude variables which did
not differ significantly across the three resolutions. Third, for the
remaining variables, we use logit analyses on all possible two and
three groups to test the significance of these variables and select
five accounting variables and five non-accounting variables that are
statistically significant in at least one combination. Since
liquidation is the worst scenario and acquisition is the best, we
define the dependent variable: resolution as 0,1 and 2 for acquired,
emerged and liquidated firms, respectively. The following ten
variables are used in this study.

(a) Five Financial Accounting Variables

NI/TA ˆ Net Income/Total Assets. This variable is used by
Ohlson (1980) and other studies. We expect a
negative estimated coefficient in the logistic
regression since as this variable increases, the
probability of liquidation decreases.
LNTA ˆ The natural logarithm of (Total Assets/GDP
deflator). This variable is a proxy for firm size
and is suggested by Ohlson (1980). We expect a
negative estimated coefficient in the logistic
regression since as size increases, the probability
of liquidation decreases.
INTA/S ˆ Intangible Assets/Net Sales. We expect a negative
estimated coefficient in the logistic regression
since as intangible assets increase, the firm
becomes a more attractive target for acquisition.
TD/TL ˆ Interest Bearing Debt/Total Liabilities. We expect
a positive estimated coefficient in the logistic
regression since as this variable increases, the
probability of liquidation increases.

ß Blackwell Publishers Ltd 2002

508 BARNIV, AGARWAL AND LEACH

SED/TD ˆ Secured Interest Bearing Debt/Total Liabilities.

We expect a positive estimated coefficient in the
logistic regression since as this variable increases,
the probability of liquidation also increases
because a secured-debt claimant would tend to
promote liquidation.

(b) Five Non-Financial Accounting Variables

FRAUD ˆ Fraudulent Activity (1 if fraud is reported, 0 other-
wise). We expect a positive estimated coefficient in
the logistic regression since as this variable
increases, the probability of liquidation increases.
RESN ˆ Resignation by Top Management (1 if resignation
of top management is reported, 0 otherwise). We
expect a negative estimated coefficient in the
logistic regression, since increase in this variable
implies new management, which would decrease
the probability of liquidation.
C-DEBT ˆ The number of major classes of bond holders.
Based on the univariate analysis, we expect a
negative estimated coefficient in the logistic
regression since as the number of classes of bond
holders increases, the probability of liquidation
decreases.
H-H INDX ˆ Herfindahl-Hirchman Index of competition. The
index varies between 0 and 1. We expect a positive
estimated coefficient in the logistic regression
since as this variable increases, the probability of
liquidation increases because there is more
competition in the industry.
PLOSS ˆ Price weighted CARs from 60 days prior to filing
through one day after the filing. We expect a
negative estimated coefficient in the logistic re-
gression since as this variable increases (i.e., more
positive or less negative returns), the probability of
liquidation decreases. Since market prices for
several bankrupt firms are very low, the CAR for
each firm is weighted by a compound price change
index, suggested by Mikkelson and Ruback (1985).

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 509

5. EMPIRICAL RESULTS

(i) Univariate Analyses

Table 2 presents univariate comparisons and tests of significance
for the ten variables used in our analysis. Panel A presents the
sample means, Panel B shows the t -tests and Panel C exhibits the
Wilcoxon Z-tests. For the accounting variables, the results suggest
that LNTA is significantly larger for acquired and emerged firms
compared with liquidated firms. INTA/S is significantly larger for
acquired firms. This result is not surprising since more intangible
assets makes firms more attractive targets for acquisition. Most of
the differences between the other three accounting variables are
not statistically significant.
The comparisons of the means for the non-accounting
variables are statistically significant for four of the five variables.
FRAUD is significantly larger for liquidated firms. C-DEBT is
significantly smaller for the liquidated firms. The H-H INDX is
significantly greater for the liquidated firms. Liquidated firms
demonstrate statistically significant higher PLOSS. The
differences across the resolutions for RESGN are statistically
insignificant. In sum, acquired firms are medium-size companies,
with greater intangible assets. Emerging firms are larger, tend to
have less secured debt and more classes of bondholders.
Liquidated firms are smaller, tend to have more secured debt
and are more subject to fraud.

(ii) Logistic Regressions

Table 3 provides the ten-variable logistic regression results for the
entire sample. Two columns are presented in the table. The first
column includes all the three groups for the entire sample. The
mean estimated coefficients based on the Lachenbruch
technique are presented. The second column provides the mean
estimated Lachenbruch's coefficients for the two-group model
for acquired and emerged vs. liquidated firms. The logistic
regressions are highly significant and the Pseudo-R2 vary between
0.11 and 0.20. The estimated coefficients for LNTA and SED/TD
are statistically significant across the two columns. These results
indicate that the smaller the firm and higher the proportion of
secured debt, the higher the probability that the firm would be

ß Blackwell Publishers Ltd 2002

 510
Table 2
Univariate Comparisons of Variables and Tests of Significance

Variables
Accounting Variables Non-Accounting Variables
Group NI/TA LNTA INTA/S TD/TL SED/TD FRAUD RESN C-DEBT H-H PLOSS
INDEX

BARNIV, AGARWAL AND LEACH

Panel A: Sample Means
Aquired ÿ0.70 3.68 0.14 0.62 0.28 0.04 0.24 5.10 0.08 ÿ0.483
Emerged ÿ0.33 3.81 0.03 0.58 0.24 0.06 0.25 5.89 0.10 ÿ0.569
Liquidated ÿ0.26 2.85 0.04 0.54 0.32 0.28 0.23 3.77 0.10 ÿ0.657

Panel B: t-tests
Acq v. Liq. ÿ1.06 2.38a 1.59 2.13b ÿ0.73 ÿ3.83a 0.16 2.54a ÿ1.45 2.68a
Emr v. Liq. ÿ1.03 3.90a ÿ0.38 1.13 ÿ0.21 ÿ3.71a 0.31 4.44a ÿ0.40 2.20b
Acq v. Emr ÿ0.90 ÿ0.41 1.73b 1.25 1.03 ÿ0.47 ÿ0.10 ÿ1.28 ÿ1.36 1.34

Panel C: Wilcoxon Z-tests

ß Blackwell Publishers Ltd 2002

Acq v. Liq. ÿ0.91 3.04a 1.85b 2.08b ÿ0.69 ÿ3.26a 0.16 2.12b ÿ2.81a 2.79a
Emr v. Liq. 0.21 2.98a 0.38 1.19 ÿ1.98b ÿ4.13a 0.31 3.43a ÿ5.34a 1.97b
Acq v. Emr ÿ0.85 ÿ0.01 1.77b 1.28 1.16 ÿ0.47 ÿ0.10 0.86 ÿ2.11b 1.17

Notes:
a
Significant at p < 0:01, one-tailed test.
b
Significant at p < 0:05, one-tailed test.
 PREDICTING BANKRUPTCY RESOLUTION 511

Table 3
Ten-Variable Logistic Regression Estimates for the Entire Sample

Acq vs. Emr vs Liq Acq & Emr vs. Liq

Mean Lachenbruch Estimates*

Variables
Intercept 1 2.093 ÿ0.213
(3.50)a
Intercept 2 ÿ0.549
(ÿ0.95)
NI/TA 0.368 0.689
(1.10) (1.53)
LNTA ÿ0.230 ÿ0.314
(2.28)b (ÿ2.33)b
INTA/S ÿ1.414 ÿ0.744
(ÿ2.03)b (ÿ0.85)
TD/TL ÿ1.548 ÿ1.175
(ÿ2.07)b (ÿ1.22)
SED/TD 1.157 1.671
(1.91)b (2.17)b
FRAUD 1.864 1.987
(3.94)a (3.80)a
RESN ÿ0.669 ÿ0.740
(ÿ2.07)b (ÿ1.72)b
C-DEBT 0.016 ÿ0.109
(0.36) (ÿ1.47)
H-H INDEX 3.381 3.500
(1.78)b (1.49)
PLOSS ÿ1.118 ÿ1.350
(ÿ2.48)a (ÿ2.03)b

Log-likelihood 2 54.50a 57.48a

Pseudo-R2 0.111 0.201
Somers D 0.418 0.586

Notes:
* The mean Lachenbruch estimated coefficients yield similar coefficients to those based
on logit model for the entire sample (without repetitions).
t -tests are in parentheses.
a
Significant at p < 0:01, one-tailed test.
b
Significant at p < 0:05, one-tailed test.

liquidated. The estimated coefficients for NI/TA are not

statistically significant. The two remaining accounting variables
have significant estimated coefficients only for the three-group
analysis.

ß Blackwell Publishers Ltd 2002

512 BARNIV, AGARWAL AND LEACH

Among the non-accounting variables, the estimated

coefficients for FRAUD are highly statistically significant. For
example, for the three-group column, the t -test is 3.94 and for
the acquired and emerged versus liquidated, it is 3.80. The
estimated coefficients of RESN and PLOSS are also statistically
significant across the two columns. The estimated coefficients of
H-H INDX are significant in the first column and the estimated
coefficients for C-DEBT are not statistically significant.
Table 4 presents similar results for the within estimation sample
described in Section 4 (i). The models are highly significant and
the Pseudo-R2 are 0.14 and 0.28 for the three groups and two
groups, respectively. Again, the estimated coefficients for FRAUD
are highly statistically significant for the two sub-groupings.
Estimated coefficients of several other variables such as NI/TA,
TD/TL and H-H INDX are not statistically significant.

(iii) Classification and Prediction Accuracies

Panel A of Table 5 shows the percentage of classification accuracy
using the ten-variable model for the entire sample based on the
Lachenbruch technique. The overall classification accuracies are
61.6 percent for the three groups and 75.1 percent for the two
groups. These results are significantly higher than a pro-
portionate three-group (two-group) random classification of
38.0 (58.7) percent.10 Further, classification accuracies are
reported for each resolution. Note that the classification
accuracies for the emerged and liquidated firms are reasonably
high but the accuracy is relatively low for the acquired firms.
Interpretations of these results are limited because the cutoffs are
based on minimizing the overall number of misclassifications and
not classification accuracies for each group.
The next section of Panel A shows the prediction accuracy for
the holdout sample. The cutoff points that minimize
misclassification in the estimation sample are used to classify
observations in the holdout sample. The three-group and the
two-group prediction accuracies are 48.8 percent and 69.9
percent, respectively, significantly better than the proportionate
chance predictions of 37.5 and 54.1 percent.
Panel B of Table 5 shows that the five variable non-accounting
model provides better classification than the five-variable

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 513

Table 4
Ten-Variable Logistic Regression Estimates for the Estimation Sample

Acq vs. Emr vs Liq Acq & Emr vs. Liq

Variables
Intercept 1 2.491 ÿ0.318
(2.73)a (ÿ0.25)
Intercept 2 ÿ0.510
(ÿ0.58)
NI/TA 0.098 0.118
(0.20) (0.16)
LNTA ÿ0.413 ÿ0.395
(ÿ2.73)a (ÿ1.57)
INTA/S ÿ5.609 ÿ17.322
(ÿ1.22) (ÿ1.65)
TD/TL ÿ1.773 ÿ1.949
(ÿ1.64) (ÿ1.25)
SED/TD 1.418 2.452
(1.64) (1.88)b
FRAUD 2.696 3.421
(3.42)a (3.27)a
RESN ÿ0.563 ÿ1.225
(ÿ1.03) (ÿ1.37)
C-DEBT 0.116 ÿ0.081
(1.68)b (ÿ0.67)
H-H INDEX 1.141 2.429
(0.39) (0.59)
PLOSS ÿ0.572 ÿ1.190
(ÿ0.71) (ÿ0.98)

Log-likelihood 2 32.14a 33.19a

Pseudo-R2 0.140 0.283
Somers D 0.467 0.686
Notes:
t -tests are in parentheses.
a
Significant at p < 0:01, one-tailed test.
b
Significant at p < 0:05, one-tailed test.

accounting model. Panels A and B demonstrate that for the

entire sample the classification accuracies of the ten-variable
model tend to be better than those for the five accounting
variables and five non-accounting variables models.
Table 6 shows the results of ECMs for the entire sample and the
holdout sample. The results demonstrate that the ECM for

ß Blackwell Publishers Ltd 2002

514 BARNIV, AGARWAL AND LEACH

Table 5
Percentages of Correct Classification Using Logistic Regressions

Model Acquired Emerged Liquidated Acq & Emr All Firms

% % % % %

Panel A: The Ten-Variable Model

The Entire Samplea
Three-Group
Classifications 12.2 82.3 60.8 61.6
Acquired and Emerged
vs. Liquidated 56.6 82.7 75.1
b
The Holdout Sample
Three-Group
Classifications 27.3 63.2 40.9 48.8
Acquired and Emerged
vs. Liquidated 27.3 93.7 69.9

Panel B: Three-Group Accounting Versus Non-Accounting Models

The Entire Samplea
Five-Variable Accounting
Model 10.2 67.2 42.1 48.2
Five-Variable Non-
Accounting Model 12.2 75.6 50.7 55.3
b
The Holdout Sample
Five-Variable Accounting
Model 27.3 61.4 38.7 47.2
Five-Variable Non-
Accounting Model 36.4 66.7 43.2 52.8

Notes:
a
The percentages reported are based on cutoff points that minimize the total number of
misclassifications for all firms in the sample based on the Lachenbruch technique.
b
Cutoff points that minimize the number of misclassifications in the estimation sample
are used to predict classification accuracy in the holdout sample.

acquired and emerged vs. liquidated is significantly lower (i.e.,

better) than that for the three groups. In addition, the ECMs for
the five-variable non-accounting models tend to be better than
those for the five-variable accounting model. Further compari-
sons across the models for the holdout sample indicate that
ECMs for the ten-variable models tend to be lower (better) than
the ECMs for the five-variable models, except for the ten-variable
model for the three-group classification. Thus, the results
reported in Tables 5 and 6 suggest that non-accounting

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 515

Table 6
Expected Cost of Misclassification (ECM)

Model Entire Estimation Holdout

Sample a Sample b Sample b

Panel A: The Ten-Variable Model

Three-Group Classifications 0.6118 0.5526 0.9349
Acquired and Emerged vs. Liquidated 0.5020 0.4511 0.6801

Panel B: Three-Group Accounting Versus Non-Accounting Models

Five-Variable Accounting Model 0.8607 0.7980 0.8790
Five-Variable Non-Accounting Model 0.7163 0.7456 0.8618

Notes:
a
The ECMs are based on Lachenbruch classifications reported in Table 5.
b
Cut-off points used for the estimation and holdout sample are based on those that
minimize the ECMs in the estimation sample. Therefore, results for the estimation sample
are also reported.

information improves the predictive ability of the final resolution

given that accounting information is already included in the
model.

6. SUMMARY AND CONCLUSIONS

This study classifies and predicts the court resolution following

bankruptcy filing. Prior studies concentrate on predicting the
bankruptcy filing event or discriminating between healthy and
financially distressed firms. About 17 months, on an average,
following the bankruptcy filing event, US bankruptcy courts con-
firm one of three possible final resolutions, namely, acquisition,
emergence or liquidation. Classifying and predicting the final
resolution is more difficult than discriminating between healthy
and bankrupt firms because all filing firms are already in financial
distress. It is important to predict the final resolution because of
the substantial economic consequences to different stakeholders.
For example, investors in liquidated firms sustain significant
losses prior to filing and from filing through the final resolution.
Whereas, investors in acquired and emerged firms gain significant
positive returns from the filing date through to the final
resolution, after sustaining negative returns prior to filing.

ß Blackwell Publishers Ltd 2002

516 BARNIV, AGARWAL AND LEACH

We use a sample of 237 firms of which 49 were acquired, 119

emerged from bankruptcy and 69 were liquidated. Logistic
regression is used to classify and predict the final resolution for
the three groups, namely, acquired, emerged and liquidated and
two groups, namely, acquired and emerged vs. liquidated. A ten-
variable model, which includes five accounting and five non-
accounting variables, performs quite well, correctly classifying
61.6 percent of all firms in their respective three groups and
correctly classifying 75.1 percent of the firms in the two-group
setting. These classification accuracies are based on the
Lachenbruch U-technique. In addition, an inter-temporal split
of the sample to an estimation sample and a holdout sample is
performed. The classification accuracies for the holdout sample
are 48.8 percent for the three groups and 69.9 percent for the
two groups.
There are several implications of this study for investors and
other stakeholders. The suggested models are useful for investors
in predicting the final resolution of the bankruptcy process.
These predictions are helpful for investment strategies in filing
firms. Creditors and other claimants may utilize the models while
negotiating the terms of the reorganization for filing firms.
Predicting the resolution would help bankruptcy judges in
making decisions about the final resolution. Auditors may use
the predictions when considering a going-concern or similar
qualification. Future research may examine these and other
discriminating variables for developing better predictions of the
final resolution. In addition, the better ability to predict the final
resolution could potentially reduce the bankruptcy costs, shorten
the reorganization process, and provide better control for the
various key stages during the bankruptcy process.
We find that only a few of the accounting variables used in
prior literature for predicting financial distress or bankruptcy
filing are useful for predicting the final bankruptcy resolution
event. Non-accounting variables such as fraud, resignation of
executives, and investors' past losses seem to be more useful for
investors, creditors and bankruptcy judges in predicting the post-
filing resolution. This relates to another important public policy
issue of the relevancy of accounting disclosure. The study
indicates that increasing non-accounting disclosure in financial
statements may benefit the stakeholders of bankrupt firms. For

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 517

example, disclosing non-accounting items in notes to annual

reports, petitioning reports and plan confirmation reports may
improve the information provided to various stakeholders. In
sum, it appears that greater and better accounting and non-
accounting disclosure may be necessary to provide external users
with relevant information for ascertaining the likely outcome of
Chapter 11 filing ex ante.

NOTES
1 Chapter 11 of the Bankruptcy Reform Act (1978) allows the court to extend
the longevity of the reorganization. Bankruptcy court may convert the firm
from chapter 11 to liquidation under chapter 7.
2 Classification refers to entire sample using the Lachenbruch technique
whereas prediction refers to the holdout sample.
3 Holdout returns from the filing through the final resolution were
moderately positive for acquired firms and slightly lower but positive for
emerging firms and moderately negative for liquidated firms.
4 Given the composition of the sample, the expected value of classification
accuracy for random sampling without replacement, would be 38 percent
for the three-group resolution.
5 In a different context, this finding is similar to results presented by Amir
and Lev (1996) who find that non-accounting data are more relevant than
accounting data for evaluation of the fast growing wireless communication
firms.
6 Early studies by Altman (1968), Deakin (1972), Edmister (1972) and Blum
(1974) used MDA to predict and classify bankruptcies. More recent studies
used logit or probit models to predict bankruptcy (Ohlson, 1980; Zmijewski,
1984; Zavgren, 1985; Burgstahler et el., 1989; Barniv, 1990; and Gilbert et al.
1990). Studies also concentrate on selecting financial variables and
comparing estimation procedures for predicting bankruptcy (Frydman et
al., 1985; Palepu, 1986; and Tam and Kiang, 1992). Taffler (1984) and
Keasey and McGuinness (1990) examine financial distress in the UK while
Micha (1984) and Skogsvik (1990) examine business failures in France and
Sweden, respectively. Altman (1984) provides a review of failure prediction
models in various countries.
7 Alternative techniques such as Nonparametric Discriminant Analysis
(NPDA) described by Barniv and Raveh (1989), and NN described for
example, by Tam and Kiang (1992), are also utilized but the results are less
defensible so they are not reported here. For instance, Trigueiros and
Taffler (1996) demonstrate some inappropriateness of such intensive
computer algorithms for statistical analysis of this nature.
8 We exclude utility, railroad, health-care, trucking and other transportation
and other regulated industries such as financial institutions including
banks, insurance companies, mortgage and thrift institutions. We exclude
these firms because the bankruptcies are handled differently for these
industries. Firms that filed for bankruptcy more than once and companies
which initially filed for Chapter 7 are also excluded.

ß Blackwell Publishers Ltd 2002

518 BARNIV, AGARWAL AND LEACH

9 The list of scrutinized variables includes some characteristics mentioned in

prior bar deliberations and by other sources such as Dun and Bradstreet.
Ten of these variables are included in our final analysis. Our analysis
excludes variables such as firm age, ownership concentration, debt-maturity
dates, which are eliminated through the selection process described in the
next paragraph.
10 The proportionate random (chance) criterion for the entire sample is
(492 ‡ 1192 ‡ 692)/2372 ˆ 0.380 for the three-group classification and
(1682 ‡ 692)/2372 ˆ 0.587 for the two-group classification.

REFERENCES
Akhigbe, A. and J. Madura (1996), `Intra-Industry Effects of Voluntary
Corporate Liquidations', Journal of Business Finance & Accounting, Vol. 23
(September), pp. 915±30.
Altman, E.I. (1968), `Financial Ratios, Discriminant Analysis, and the Prediction
of Corporate Bankruptcy', The Journal of Finance, Vol. 23 (September), pp.
589±609.
__________ (1984), `The Success of Business Failure Prediction Models: An
International Survey', Journal of Banking and Finance, Vol. 8 (June), pp. 171±
98.
Amir, A. and B. Lev (1996), `Value-Relevance of Non-financial Information:
The Wireless Communication Industry', Journal of Accounting and Economics,
Vol. 22 (January), pp. 3±30.
Barniv, R. (1990), `Accounting Procedures, Market Data, Cash-Flow Figures and
Insolvency Classification: The Case of the Insurance Industry', The
Accounting Review, Vol. 65 (July), pp. 578±604.
__________ and A. Raveh (1989), `Identifying Financial Distress: A New
Nonparametric Approach', Journal of Business Finance & Accounting, Vol. 16
(Summer), pp. 361±83.
Blum, M. (1974), `Failing Company Discriminant Analysis', Journal of Accounting
Research, Vol. 12 (Spring), pp. 1±25.
Burgstahler, D., J. Jiambalvo and E. Noreen (1989), `Changes in the Probability
of Bankruptcy and Equity Value', Journal of Accounting and Economics, Vol.
11 (No. 3), pp. 207±24.
Casey, C., V. McGee and C. Stickney (1986), `Discriminating Between
Reorganized and Liquidated Firms in Bankruptcy', The Accounting Review,
Vol. 6 (April), pp. 249±62.
Deakin, E. (1972), `A Discriminant Analysis of Predictors of Business Failure',
Journal of Accounting Research, Vol. 10 (Spring), pp. 167±79.
Dhumale, R. (1998), `Earning Retention as Specification Mechanism in Logistic
Bankruptcy Models: A Test of the Free Cash Theory', Journal of Business
Finance & Accounting, Vol. 25 (September/October), pp 1005±24.
Dopuch, N., R.W. Holthausen and R.W. Leftwich (1987), `Predicting Audit
Qualifications with Financial and Market Variables', The Accounting Review,
Vol. 62 (July), pp. 431±54.
Edmister, R.O. (1972), `An Empirical Test of Financial Ratio Analysis for Small
Business Failure Prediction', Journal of Financial and Quantitative Analysis,
Vol. 7 (March), pp. 1477±93.
Frydman H., E.I. Altman and D. Kao (1985), `Introducing Recursive

ß Blackwell Publishers Ltd 2002

 PREDICTING BANKRUPTCY RESOLUTION 519
Partitioning for Financial Classification: The Case of Financial Distress',
The Journal of Finance, Vol. 40 (March), pp. 269±91.
Gilbert, L.R., K. Menon and K.B. Schwartz (1990), `Predicting Bankruptcy for
Firms in Financial Distress', Journal of Business Finance & Accounting, Vol.
17, No. 1 (Spring), pp. 161±71.
Hotchkiss, E.S. (1995), `Postbankruptcy Performance and Management
Turnover', Journal of Finance, Vol. 50 (March), pp. 3±21.
Hsieh, S.J. (1993), `A Note on the Optimal Cutoff Point in Bankruptcy
Prediction Models', Journal of Business Finance & Accounting, Vol. 20 (April),
pp. 457±64.
Keasey, K. and P. McGuinness (1990), `The Failure of UK Industrial Firms for
the Period 1976-1984, Logistic Analysis and Entropy Measure', Journal of
Business Finance & Accounting, Vol. 17 (Spring), pp. 119±35.
Kim, M. and M. Kim (1999), `A Note on the Determinants of the Outcomes of
Bankruptcy Petitions: Evidence From Korea', Journal of Business Finance &
Accounting, Vol. 26 (September/October), pp. 997±1011.
Lachenbruch, P.A. (1967), `An Almost Unbiased Method for Obtaining
Confidence Intervals for the Probability of Misclassification in Discriminant
Analysis', Biometrics, Vol. 23 (December), pp. 639±45.
Laitnen, E.K. and T. Laitnen (1998), `Cash Management Behavior and Failure
Prediction', Journal of Business Finance & Accounting, Vol. 25 (September/
October), pp. 843±920.
Lau, A. Hing-Ling (1987), `A Five-State Financial Distress Prediction Model',
Journal of Accounting Research, Vol. 25 (Spring), pp. 127±38.
Lennox, C.S. (1999), `The Accuracy and Incremental Information Content of
Audit Reports in Predicting Bankruptcy', Journal of Business Finance &
Accounting, Vol. 26 (June/July), pp. 757±778.
Liang,T.P., J.S. Chandler, I. Han and J. Roan (1992), `An Empirical
Investigation of Some Data Effects on the Classification Accuracy of Probit,
ID3, and Neural Networks', Contemporary Accounting Research, Vol. 9
(Spring), pp. 306±28.
LoPucki, L.M. (1983), `The Debtor in Full Control-System Failure Under
Chapter 11 of the Bankruptcy Code?', American Bankruptcy Law Journal, pp.
99±126.
Maddala, G.S. (1983), Limited Dependent and Qualitative Variables in Econometrics
(Cambridge University Press, Cambridge).
__________ (1991), `The Perspective on the Use of Limited-Dependent and
Qualitative Variables Models in Accounting Research', The Accounting
Review, Vol. 66 (October), pp. 788±807.
Micha, B. (1984), `Analysis of Business Failure in France', Journal of Banking and
Finance, Vol. 8 (June), pp 281±92.
Mikkelson, W.H. and R.S. Ruback (1985), `An Empirical Analysis of the
Interfirm Equity Investment Process', Journal of Financial Economics, Vol. 14
(No. 4), pp. 523±53.
Morse, D. and W. Shaw (1988), `Investing in Bankrupt Firms', Journal of Finance,
Vol. 43 (December), pp. 1193±206.
Ohlson, J.A. (1980), `Financial Ratios and the Probabilistic Prediction of
Bankruptcy', Journal of Accounting Research, Vol. 18 (Spring), pp. 109±31.
Palepu, K.G. (1986), `Predicting Takeover Targets: A Methodological and
Empirical Analysis', Journal of Accounting and Economics, Vol. 8, No. 1, pp. 3±
35.
Platt. H.D. and M.B. Platt (1990), `Development of a Class of Stable Predictive

ß Blackwell Publishers Ltd 2002

520 BARNIV, AGARWAL AND LEACH

Variables: The Case of Bankruptcy Prediction', Journal of Business Finance &

Accounting, Vol. 17 (Spring), pp. 31±51.
Platt, H.D., M.B. Platt and J.G. Pedersen (1994), `Bankruptcy Discrimination
with Real Variables', Journal of Business Finance & Accounting, Vol. 21 (June),
pp. 491±510.
Richardson, F.M, G.D. Kane and P. Lobingier (1998), `The Impact of Recession
on the Prediction of Corporate Failure', Journal of Business Finance &
Accounting, Vol. 25 (January/March), pp. 167±86.
Ro, B.T., V. Zavgren and S.J. Hsieh (1992), `The Effect of Bankruptcy on
Systematic Risk of Common Stock: An Empirical Assessment', Journal of
Business Finance & Accounting, Vol. 19 (April), pp. 309±28.
Skogsvik, K. (1990), `Current Cost Accounting Ratios as Predictors of Business
Failures: The Swedish Case', Journal of Business Finance & Accounting, Vol.
17 (Spring), pp. 137±60.
Taffler, R.J. (1984), `Empirical Models for Monitoring of UK Corporations',
Journal of Banking and Finance, Vol. 8 (June), pp. 199±227.
Tam, K.Y. and M.Y. Kiang (1992), `Managerial Applications of Neural Networks:
The Case of Bank Failure Predictions', Management Science, Vol. 38 (July),
pp. 926±47.
Tennyson, B.M, R.W. Ingram and M.T. Dugan (1990), `Assessing the
Information Content of Narrative Disclosures in Explaining Bankruptcy',
Journal of Business Finance & Accounting, Vol. 17 (Summer), pp. 391±410.
Trigueiros D. and R. Taffler (1996), `Neural Networks and Empirical Research
in Accounting', Accounting and Business Research, Vol. 26, No. 4, pp. 347±55.
Ward, T.J. (1994), `An Empirical Study of the Incremental Predictive Ability of
Beaver's NaõÂve Operating Flow Measure Using Four-State Ordinal Models
of Financial Distress', Journal of Business Finance & Accounting, Vol. 21
(June), pp. 547±61.
__________ and B.P. Foster (1997), `A Note on Selecting a Response Measure
for Financial Distress', Journal of Business Finance & Accounting, Vol. 24
(July), pp. 869±78.
White, M.J. (1983), `Bankruptcy Costs and the New Bankruptcy Code', Journal of
Finance, Vol. 38 (June), pp. 477±87.
__________ (1989), `The Corporate Bankruptcy Decision', Journal of Economic
Perspectives, Vol. 3 (Spring), pp. 129±51.
Zavgren, C. (1985), `Assessing the Vulnerability to Failure of American
Industrial Firms: A Logistic Analysis', Journal of Business Finance &
Accounting, Vol. 12 (Spring), pp. 19±45.
Zmijewski, M.E. (1984), `Methodological Issues Related to the Estimation of
Financial Distress Prediction Models', Journal of Accounting Research, Vol. 22
(Spring), pp. 59±82.

ß Blackwell Publishers Ltd 2002