Journal of Banking and Finance 8 (1984) 171-198.

North-Holland

THE SUCCESS OF BUSINESS FAILURE PREDICTION MODELS

An International Survey*

Edward I. ALTMAN
New York University, New York, NY 10006, USA

This paper surveys and discusses numerous studies, both published and unpublished, that have
attempted to construct and test business failure models outside the United States. Failure risk
models are one of the few types of financial models that have been applied pervasively on an
international level. A considerable flow of literature has gone relatively unnoticed due to the fact
that the reference source is not well known or the source material is not available in English. It
is the purpose of this paper to document these efforts and thereby increase the discourse and
knowledge about such studies and others not surveyed here. This survey complements the other
company related studies found in this Special Issue.

1. Introduction and background

Business failure identification and early warnings of impending financial

crisis are important not only to analysts and practitioners in the United
States. Countries throughout the world, even non-capitalist nations, are
concerned with individual firm performance assessment. Developing countries
and smaller economies, as well as the larger industrialized nations of the
world, are vitally concerned with avoiding financial crises in the private and
public sectors. Smaller nations are particularly vulnerable to financial panics
resulting from failures of individual entities.
In the late 1960s and throughout the 197Os, numerous studies in the
United States were devoted to assessing one’s ability to combine publicly
available data with statistical classification techniques in order to predict
business failure. A steady stream of failure prediction papers have appeared
in the English literature, and numerous textbooks and monographs include a
section or chapter on these models. What has gone relatively unnoticed, on
an international level, is the considerable efforts made to replicate and extend
these models to environments outside the U.S. There is no work with which

*This is a shortened version of a monograph, Occasional Paper no. 5, 1982, Salomon Brothers
Center for the Study of Financial Institutions, NYU. The author would like to express his
appreciation to Jay Ko (New York University Graduate School of Business Administration), K.
Takahashi (Keio Business School) and J.H. von Stein (Hohenheim University, Germany) for
providing translated material which was indispensable for this survey. In addition, several of the
authors cited in this paper provided valuable unpublished materials and feedback which added
to the work’s overall com.pleteness.

037884266/84/$3.00 0 1984, Elsevier Science Publishers B.V. (North-Holland)

172 E.I. Altman, Success of business failure models

we are familiar that attempts to survey these studies and to comment on

their similarities and differences. The purpose of this analysis is to do just
that. We survey the works by academics and practitioners in 10 countries
and give references to several other studies. In addition, background data on
recent bankruptcy rates in several of the major countries of the world are
compared with Rates in the U.S. We also examine financial ratios and failure
indicators (Z-scores) on an international basis.
In many cases, I can present an in-depth discussion of the models
including individual variable weights. In others, I present the models in more
general terms due to the lack of precise documentation. Still, we feel that this
international treatment of failure prediction models is the most
comprehensive effort to date. I recognize that some relevant works will
possibly be overlooked in this survey and apologize for any omission.
Since the late 1960s there has been considerable interest among researchers
in the development and testing of models for classifying and predicting
business failures. Probably the two most influential works were by Beaver
(1966) and Altman (1968). Both presented methodologies that have been
replicated and improved on for many different types of firms and in a
number of foreign environments. Beaver segregated for analysis a sample of
bankrupt firms and a matched sample of non-bankrupt firms and studied the
2 samples’ financial performance indicators for up to 5 years before failure.
Beaver’s work was a type of univariate analysis whereby each measure, or
ratio, was analyzed separately and the optimal cutoff point was selected so
that the number of accurate classifications was maximized for that particular
sample. When his analysis is evaluated based on the original samples, the
technique is known as classification analysis. When a model is tested on a
sample of firms other than the original one, preferably from a period after
the original model’s data source, the analysis takes on a predictive flavor.
Beaver tested 14 ratios and found that the cash flow to total debt ratio
was the best classifier of corporate bankruptcy. Other important financial
measures were the debt to total assets and net income to total assets ratios,
and the ‘no credit interval’ [for precise definitions, see Beaver (1966)].

The Z-score model. Altman was the first to apply the technique known as
discriminant analysis to the failure classification problem. The analysis is
multivariate in that a number of variables are combined simultaneously to
analyze a firm for its failure potential. That particular technique, known as
Z-score model, relevant for manufacturing entities only, has been applied by
practitioners to problems of credit analysis, investment analysis, and going-
concern evaluation, among others. The Z-score model is expressed as follows:

Z = 1.2X, + 1.4X, +3.3X, +0.6X, +0.99X,,

 EL Altman, Success of business failure models 173

where
X, = (current assets -current liabilities)/total assets,
X, = retained earnings/totals assets,
X, =earnings before interest and taxes/total assets,
X, = market value of preferred and common equity (number of shares x price
of stock)/total liabilities,
X, = sales/total assets.

Any firm with a Z-score below 1.8 is considered to be a prime candidate

for bankruptcy, and the lower the score, the higher the failure probability.
This model was over 90% accurate in classifying bankrupt firms correctly
one statement prior to failure and over 80% accurate in subsequent prediction
tests. We will return to the Z-score analysis at several points later in this
survey.
With the many important changes in reporting standards since the 1960s
the Z-score model is somewhat out of date in the 1980s. A second-generation
model known as Zeta analysis [Altman et al. (1977)] adjusts for these changes,
primarily the capitalization of financial leases. The resulting linear Zeta
discriminant model is extremely accurate for up to 5 years before failure.
Since this analysis is a proprietary one, the exact weights for the model’s
seven variables cannot be specified here.

Failure prediction services. There are at least 4 statistical ‘services’ in the

U.S. that seek to assess the insolvency risk of industrial companies:

(1) The Z-score model. This model is in the public domain, and it is not
necessary to subscribe to a statistical service, although some firms, such as
Merrill Lynch, have provided Z-scores.
(2) The Zeta model. Available from Zeta Services, Inc., Mountainside, New
Jersey.
(3) The gambler’s ruin model. Developed by Wilcox (1971,1976). Available
from Advantage Financial Systems, Boston.
(4) The QES score. Available from the Trust Division of the First Union
Bank, Charlotte, North Carolina.

2. Japan

2.1. Background data and comparative analyses

Despite its exceptional economic performance in the last decade, the

Japanese economy, and its component firms, suffers from the same business
failure realities as do the western economies. The number of business failures,
prior to 1982 in Japan, is at least as large as in the U.S. on a relative basis.
174 E.I. Altman, Success of business failure models

Table 1 compares the bankruptcy rate in the two countries. Except for the
most recent 2 years (1982-1983), the U.S. had a lower rate based on GNP
and population bases and a higher rate based on a per-firm base. Since the
U.S. business bankruptcy figures include a large number of non-incorporated
firms, it is also likely that this latter comparison (for corporations alone)
would show the Japanese rate to be relatively higher; a fact that will surprise
many observers.

Table 1
Relative business bankruptcy rates, U.S.A. and Japan, 1976-1981.”

1976 1977 1978 1979 1980 1981

Bankruptcies
U.S.A. 35.201 32,189 30,528 29,500 36,411 47,414
Japan 15,641 18,741 15,875 16,030 17,884 17,610
Ratio 2.25 1.72 1.92 1.84 2.04 2.69

Populatzon (million)
U.S.A. 215.1 216.8 217.0 218.2 226.5
Japan 113.1 114.2 115.2 116.3 117.0 -
Ratio 1.90 1.90 1.88 1.88 1.94

Gross national product ($U.S billion)

U.S.A 1706.5 1889.6 2127.6 2368.3 2730.6
Japan 565.9 693.5 974.5 1011.2 1096.0
Ratio 3.01 2.73 2.18 2.34 2.49 -

Corporations (million)
U.S.A 2024 2241 2410 2520 2605” -
Japan 1293 1351 1426 1494 1570
Ratio 1.56 1.66 1.69 1.69 1.66 -

“Sources: U.S. Administrative Courts, Bankruptcy Division, Statistical

Abstract of the United States; Tokyo Shoko Koshinsho, Oriental Economist,
Statistics Monthly; Industrial Review of Japan.
bEstimate.

In March of 1984, J. Osawa & Company, a trading concern, filed for court
protection under the Japanese bankruptcy laws. Osawa was the third largest
Japanese business failure in the post world war period with liabilities
exceeding the equivalent of $500 million. The two largest bankruptcies
involved the Sanyo Tokushuko Corporation, a specialty steel maker, and
Kojin, a chemicals manufacturer. In Japan, as in the U.S. today, no longer is
the large firm invulnerable to financial distress although such large failures
are less frequent in Japan due to protection from industrial groups with big
banks at the financial center. Osawa was not a member of such a group and
its bankruptcy caused a number of other companies also to fail.
In Japan, bankruptcies are concentrated in the small and medium-sized
firms, especially those that do not enjoy the protection of an affiliated group
 E.I. Altman, Success of business failure models 17.5

of companies. These groups, known as ‘Keiretsu’, usually involve a lead

commercial bank and a number of firms in diverse industries. Still, a number
of larger firms listed on the first section of the Tokyo Stock Exchange have
succumbed to the negative economic reality of failure. These firms,
numbering close to 50 over the period 1963-1978, provide a sample of firms
sufficiently large enough for rigorous statistical analysis. Table 2 lists the
causes of bankruptcies in Japan and the U.S. It is interesting to note that in
Japan a significant percentage of failures are caused by other firms’ problems,
which illustrates the complex interactions among companies. Of course, the
categories selected by two of the leading bankruptcy statistics publishers,
Dun & Bradstreet and Tokyo Shoko Research Ltd., are not the same.

Table 2
Listed categories of business failure causes in U.S.A. and Japan, 1980.”

Cause Percent of failures (1978)

U.S.A.
Managerial incompetence 47.1
Lack of managerial experience 27.0
Unbalanced experience 17.9
Neglect 1.0
Fraud 0.5
Unknown 6.5

Total 100.0

Slowdown of sales 41.5

Irresponsible management 23.0
Directly due to other bankruptcies 11.7
Indirectly due to other bankruptcies 8.1
Under capitalization 5.8
Bad debts 3.7
Excessive equipment investments 3.3
Other 2.9

Total 100.0

“Sources: Failure Record, Dun & Bradstreet (1980), Tokyo Shoko

Koshinsho, Oriental Economist, Statistics Monthly.

A number of Japanese academics have attempted to analyze the business

failure classification problem. One study, Takahashi et al. (1979) will be
discussed in detail in this issue. A summary of other studies is provided in
Altman (1983).
The Z-score model was applied to a large number of Japanese entities by
the economic-business publication Nikkei-Business (1978). In this article, it
was found that approximately 9% of the population of Japanese firms
examined (90 of slightly under 1000 firms) had Z-scores below the lower
176 EL Altman, Success of business failure models

bound of the ‘Japanese cutoff-score’. The cut-off-score was subjectively

adjusted by the author of the article to account for the unique features of the
Japanese environment. Essentially, the adjustments considered the difference
in accounting treatment for retained earnings and for the higher fixed debt
carried by firms in Japan relative to U.S. firms. The lower bound cutoff-score
was lowered from Altman’s 1.81 to 1.0 for Japanese companies. I cannot
comment on the validity of the 1.0 score but I agree that the relevant cutoff-
score is probably lower in Japan, due primarily to the factors cited above
and a lower average return on assets for Japanese firms.
If one would simply plug the values for the 5 factors of the Z-score model
for the average bankrupt firm in Japan, the resulting average bankrupt firms’
score would be 0.67 compared to the average U.S. firms’ score of -0.27,
based on Altman’s (1968) sample. But, we have noted that every firm in the
original sample whose Z-score was below 1.81 failed. In this context, Nikkei-
Business’ cutoff-score of 1.0 is realistic. We will explore the differences in
average Z-score for a number of countries at a later point (table 5).

2.2. The Ko (1982) analysis of problemJirms

In addition to the models developed in Japan, Ko (1982) has also analyzed

Japanese problem firms. His sample included 41 pairs of bankrupt and non-
bankrupt entities for the period 1960-1980. Several accounting corrections,
adjustments, and transformations, in addition to variable trends, were
applied to the data-set to reduce the biases held to be inherent in
conventional Japanese reporting practices. He compared the standard linear
model design against a linear model with first-order interactions and also
against a quadratic model. He also examined a discriminant model using
factor analysis for orthogonal variable transformation. On the basis of
classification results, a 5 variable linear independent model, without the
orthogonal transformations, was selected as the best model which yielded a
82.9% correct classification rate by Lachenbruch (1967) tests v. a 90.8% for
the original sample set. It is interesting to note that the linear interaction
design appeared best on the basis of group separations potential, but not for
classification accuracy.
Ko found, with respect to the variables of the model, that each sign was in
agreement with each variable’s economic meaning and that 3 of the variables
are similar to those in Altman’s (1968) model. They are: EBIT/sales, working
capital/total debts, and market equity/total debts. A fourth variable in this
model is an inventory turnover change ratio. His last ratio was the standard
deviation of net income over 4 periods.
The final standardized coefficient model is of the form

Zj =0.868X, +0.198X, -0.048X, +0.436X, +0.436X, +0.115X,,

where
 E.I. Altman, Success of business failure models 117

X, = EBIT/sales,
X, =inventory turnover 2 years prior/inventory turnover 3 years prior,
X, =standard error of net income (4 years),
X, = working capital/total debt,
X, = market value equity/total debt,
Zj = Z-score (Japanese model).
The standardized form results in a zero cutoff-score, i.e., any score greater
than zero indciates a healthy situation, with probability of classification of
bankruptcy less than 0.5, and probabilities greater than 0.5 for negative
scores.

2.3. A summary of Japanese bankruptcy models

At a recent conference in Keio Graduate School of Business
Administration, a summary of most of the efforts to date to build a Japanese
bankruptcy classification model was presented [Takahashi and Altman
(1981), summarized in Altman (1983, ch. 12)]. I am impressed at the number
and obvious serious attention given to this subject in Japan. This research
represents efforts by university professors and practitioners and at least 2 of
those models are presently being tested or marketed by Japanese firms.
All but two of the models utilize discriminant analysis as the statistical
classification technique. The others utilize regression and principal
component analyses. A few of the studies were concerned with smaller, non-
listed firms although the majority included only listed companies whose data
are more easily available. Despite these many efforts, the research in the
bankruptcy area is likely to continue, especially if the Japanese domestic
economy suffers a serious recession and the number of failures increases
dramatically.

3. Federal Republic of Germany (and Switzerland)

During the decade of the 1970s West Germany enjoyed exceptional
economic growth. Even during these prosperous times, the number of
business failures was relatively high reflecting a highly competitive
environment. The years 198&1982 brought significant economic pressures to
the overall economy and with them the number of failures increased by more
than 15% in 1980 and over 25% in 1981 and 1982. Insolvencies have risen
over 13% and if this pace continued, the total number for 1983 would be
over 12,000 compared to about 9,000 in the 1976-1980 period. Table 3
compares the recent bankruptcy rates for West ,Germany and the U.S.A. Just
as we discovered in the comparison of Japanese and U.S.A. rates, the West
Germany-U.S.A. ratios are quite similar. In recent years the ratio of
bankruptcies to population in both countries was about 0.00014 and the
ratio of bankruptcies to GNP was about 0.000011.
178 E.I. Altman, Success of business failure models

Table 3
Relative business bankruptcy rates, U.S.A. and West-Germany, 1976-1980.

1976 1977 1978 1979 1980

Bankruptcies
U.S.A. 35,201 32,189 30.528 29.500 36,411
Germany 9,362 9,562 8,722 8,319 9,560
Ratio 3.76 3.37 3.50 3.55 3.80

Population (million)
U.S.A. 215.1 216.8 217.0 218.2 226.5
Germany 61.5 61.4 61.3 61.4 61.5
Ratio 3.50 3.53 3.54 3.55 3.68

Gross national product ($U.S. billion)

U.S.A. 1706.5 1889.6 2127.6 2368.8 -
Germany 446.9 515.7 641.1 761.0 -
Ratio 3.82 3.66 3.32 3.11 -

“Sources: Annual Report of the Director of the &ministrative Office of the U.S.
Courts, Statistical Abstract of the U.S., Statistisches Jahrbuch der Bunderrepublik
Deutschland 1980.

3.1. An early bankruptcy classification study (Switzerland)

While bankruptcy classification and its many implications have interested
researchers in Germany for many years, the earliest major work published in
Germany was performed in Switzerland by Weibel (1973). He constructed a
sample of 36 failed Swiss firms in the 196&1971 period and matched them to
a like number of non-failed firms in terms of age, size, and line of business.
Using univariate statistical parametric and non-parametric tests, Weibel
analyzed ratios of these 2 groups in much the same way as Beaver (1966)
did. He found that many of the individual ratios were non-normal and so he
abandoned multivariate tests. [We have often referred [Altman et al. (1977)]
to the non-normality problem which exists in many economic and financial
data-sets but we prefer to test the robustness of models using such data
rather than abandoning the tests. We do observe that some European
researchers have found multivariate studies suspect due to the non-normality
properties of financial measures.]
Out of 41 original ratios, Weibel selected 20 for dichotomous comparisons.
He utilized cluster analysis to reduce collinearity and arrived at the
conclusion that 6 ratios were especially effective in discriminating between
the paired groups. Three ratios were types of liquidity measures with 1 (near
monetary resource assets-current liabilities/operating expenditures before
depreciation) performing best. He also found that inventory turnover and
debt/asset ratios were good individuai predictors. He examined the
overlapping range of individual ratios for the 2 groups and presented some
ad-hoc rules for identifying failures. He then divided the observations into 3
 E.I. Altman, Success of business failure models 119

risk groups. The low risk group had all 6 ratios in the interdecile range of
good firms; high risk firms had at least 3 ratios in the interdecile range of
failed companies; and a final category was identified where the firm does not
fall into either of the other 2 groupings. Weibel’s results were quite accurate
in the classification stage; we have no documentation on how his ‘model
performed on holdout tests and what has been the evolution of models in
Switzerland since his original work.

3.2. German studies

Many studies in Germany have investigated the causes and problems of
insolvencies, especially for financial organizations, e.g., von Stein (1968).
Beerman (1976) published one of the first German statistical classification
models for insolvency analysis. He examined matched groups of 21 firms
which operated or failed in 1966-1971. Applying dichotomous and linear
discriminant tests, he analyzed 10 ratios encompassing profitability, cash
flow, fixed asset growth, leverage, and turnover. His results, using the
difference in means dichotomous test, were mixed with one ratio type
(profitability), yielding quite respectable results. The other ratios were far less
impressive on a univariate basis. Beerman advocates using discriminant
analysis and his 10 ratio model yielded classification errors of 9.5x, 19.0x,
28.6x, and 38.1% for the 4 years prior to failure. He does not indicate which
model to use, and the coefficients of each measure were quite unstable in the
4 different year models. Also, we are given no indication of holdout test
results or predictive accuracy and, due to the small sample, we do not have
confirming evidence of the model(s) reliability.
Weinrich’s (1978) book, from his dissertation, attempted to construct risk
classes in order to predict insolvency. His sample of failed firms was
considerably larger (44) than Beerman’s, concentrating on small and
intermediate size firms, with an average sales of DM 4 million (less than $ 2
million), that failed in 1969-1975. Weinrich considered 3 consecutive annual
financial statements (years 2-4 prior to failure) but did not utilize the one
statement closest to insolvency. This is a marked difference from most of the
other models we have studied.
Weinrich abandoned the use of parametric classification techniques due to
his feeling that many assumptions were violated (normality, variance-
homogeneity of groups, and high correlation amongst the variables). His
linear discriminant models were quite good in terms of classification accuracy
(11% error for year 2, 15.7% and 21.9% for years 3 and 4, respectively).
Weinrich did use factor analysis and found the technique useful, indicating
at least 6 different factors that explained 80% of the variance of the ratios.
He then devised a model of credit worthiness that contained 8 relatively
independent ratios and utilized both univariate and multivariate methods. A
point evaluation system was devised based on quartile values of good and
bad firms. For example, a net worth/debt ratio over 43.3% receives the best
TRF-B
180 E.I. Altman, Success of business failure models

(lowest) point value. A firm with significant insolvency potential is one with
24 points or more (an average of 3 for each of the ratios). This arbitrary
point system correctly classified over 90% of the failed firms, 2 years prior to
failure. But, only 60% accurate 3 years prior. The Type II error rate was
quite high averaging well over 20% in each year. Weinrich advocated the use
of trend analysis of the point system as well as the point estimate.
Gebhardt (1980) compared dichotomous and multivariate classification
tests of samples of failed and non-failed firms based on models constructed
before and after the 1965 Financial Statement Reform Law. The earlier
model contained 13 matched pairs of industrial firms and the post-1965
model contained 28 pairs. He utilized a very large number of possible
financial indicators which were reduced to 41 ratios for the dichotomous
tests. He also incorporated crude measures of misclassification costs and
tested his results with the Lachenbruch (1967) holdout test procedure.
Gebhardt, like others, felt that the non-normality of some ratios implied the
use of non-parametric procedures but found those results unsatisfactory. The
multivariate results were far superior. Gebhardt concluded that the pre-1965
models’ results were actually better than the ones following the Reform Law
of 1965.
Fischer’s (1981) recent work concentrates on non-numerical data for
forecasting failure. He is particularly interested in methods of credit
evaluation for suppliers who do not have the ability or the data to perform
comprehensive conventional analysis on their existing and potential
customers. He advocates an electronic data processing system which can
retrieve and analyze such non-numerical information as reports from
newspapers, magazines, inquiry agencies and credit information from other
sellers. Unfortunately, according to Fischer, commercial rating agencies and
banks are constrained as to how honest and revealing they choose to be with
regard to their reports. In addition, the information provided may be
outdated and certainly contains subjective elements. More than one source of
credit information is therefore desirable.
Fischer advocates combining the permanent and transitory information on
enterprises with microeconomic and socio-political data. Five arbitrary rating
categories are devised based on non-numerical data and the delphi technique
(numerous experts in various areas) is also recommended. Each characteristic
is rated over time into the 5 categories. The sum of development patterns
from varying sources of information builds the basis for a financial
classification. Clustering techniques are also used by Fischer to clarify
information types.
The most recent attempt by von Stein is discussed in detail by the author
in this issue.

4. Brazil
Brazil is an example of an economy where the end-result of a series of
 E.I. Altman, Success of business failure models 181

economic setbacks would put severe pressure on private enterprises. For

example, tightening of credit for all firms - especially smaller ones - can
jeopardize financial institutions and undermine government efforts to
promote economic development. Most observers would agree that preventive
action to detect and avoid critical pressures of this type is highly desirable.

4.1. Evolving Brazilian Jinancial pressures

The Brazilian economy has enjoyed extraordinary growth in the last

decade despite persistently high rates of inflation. The so-called ‘Brazilian
miracle’, however, is generally believed to be finished - at least temporarily.
There have been several disastrous statistics and trends in the last few years
and these trends are expected to continue - even to worsen. Brazilian firms
in general, both national and multinational, operate with fairly high leverage
ratios. These high leverage figures mean, of course, increasing fixed financial
expenses which, in Brazil’s case, are cash outflows subject to monetary
correction. In the period 1974-1983, interest and related financial expenses
have been growing faster than earnings available to cover these expenses.
The margin of safety is diminishing dramatically and, with this drop, the
overall risk dimensions of the average Brazilian firm are a cause of concern
to managers, financial institutions, government officials, and other relevant
parties.
If one adds to the picture a relatively tight monetary policy that was
adopted in the late 1970s to combat inflation, diminishing profit margins,
and a woefully inadequate capital market - especially the lack of debt
capital sources even for the strongest firms - the result is a situation of
increasing firm pressures. The huge international debt that has been
accumulated in Brazil further exacerbates the situation.

4.2. Altman, Baidya and Riberio-Dias (1979)

Altman et al. (1979) examined two a priori groups of firms categorized as

serious-problem (S-P) and no-problem (N-P) companies. A small number of
variables were then calculated for each observation (firm) in each of the 2
samples. Data covered the period from 1 to 3 annual reporting statements
prior to the serious-problem date. The data from 1 year prior (and the
corresponding year for the control sample) were then analyzed through the
use of linear discriminant analysis.
The serious-problem firm was defined as those tiling formal petitions for
court-supervised liquidations (falencias), legal reorganizations in bankruptcy
(concordatas), and out-of-court manifestations of serious problems. In all but
2 of the 23 serious-problem cases, the serious-problem was manifest during
the 30 months, January 1975-June 1977. Industry categories represented
include textiles, furniture, pulp and paper, retail stores, plastics, metallurgy,
182 E.I. Altman, Success of business failure models

and others. The average asset size of the serious-problem firms was
surprisingly high at 323 million cruzeiros (US$ 25-30 million). Therefore, the
model, if accurate, has relevance over a wide range of companies in terms of
size. The control (or no-problem) sample was actually somewhat smaller,
with an average size of just under CR$ 300 million and a median size of
about CR$ 200 million.
One or 2 firms were selected for the control sample from each of the same
industrial categories as those represented by the serious-problem group, and
data were gathered from the year corresponding to the year prior to the
problem date. Since there were more than 30 industrial categories to choose
from, the number of firms in. each industrial group was often quite small.
Whenever possible, privately-owned, domestic companies were selected since
we felt that a state-owned or multinational affiliation precluded, in general,
the possibility of failure.
The classification procedure used in this study is based on the failure
model developed in the United States [Altman (1968)] with modifications
that allow for consideration of Brazilian standards and reporting practices.
In this Brazilian study, the same variables were utilized (see p. 4), but X, and
X, were modified. With respect to X,, the retained earnings account on U.S.
balance sheets reflects the cumulative profits of a firm less any cash dividends
paid out and stock dividends. In most instances, the small, young firm will
be discriminated against because it has not had time to accumulate its
earnings. In Brazil, however, due to different financial reporting practices and
adjustments for inflation, there is no exact equivalent to retained earnings.
The nearest translation to retained earnings is ‘lucros suspensos’ which are
those earnings retained in the business after distribution of dividends. This
amount is usually transferred, however, within a short time (for example, 2
years) through stock dividends to the account known as capital.
In addition, reserves which were created to adjust for monetary correction
on fixed assets and the maintenance of working capital were deducted from
profits and thereby decrease those earnings which are reported to be retained
in the firm. These reserves, however, increase both the assets and the firms
equity and they, too, are transferred to capital. In essence, then, that amount
of capital which represents funds contributed by the owners of the firm is the
only part of equity that is not considered in the Brazilian equivalent to
retained earnings; X, was calculated as

x = total equity-capital contributed by shareholders (CCS)

2
total assets

A more precise expression of the numerator would be the cumulative yearly

retained earnings plus the cumulative reserves created over the life of the
firm, but this information is very difficult to obtain outside the firm and was
not available to the authors.
 E.I. Ahax, Success of business failure models 183

Since most Brazilian firms’ equity is not traded, there cannot be a variable
which measures the market value of equity (number of shares outstanding
times the latest market price). To derive the new values for X,, the book
value of equity (patrimonio liquido) was substituted and divided by the total
liabilities. The remaining 3 variables were not adjusted, although we are
aware of the fact that certain financial expenses are also adjusted for inflation
in Brazilian accounting.

4.3. Empirical results

The empirical results will be discussed in terms of 2 separate, but quite
similar, models. The first model, referred to as Z,, includes variables Xz to
X5 (4 measures). Model Z1 does not include X, because the stepwise
discriminant program indicated that it did not add any explanatory power to
the model and the sign of the coefficient was contrary to intuitive logic. Once
again, as so often is found in multivariate failure classification studies, the
liquidity variable is not found to be particularly important. The second
model, referred to as Zz, does not include X, because X, is quite difficult to
derive with just one set of financial statements and it is similar to X4 Model
2, can therefore be applied without supplementary data.
The models are as follows (see earlier Z-score representation):

2, = 1.44 +4.03X, +2.25X, +0.14X, +0.42X,,

Z, = - 1.84 -0.51X, + 6.23X, +0.71X4+0.56X5.

In both cases, the critical cutoff-score is 0. That is, any firm with a score
greater than 0 is classified as having a multivariate profile similar to
continuing entities and those with a score less than 0 as having
characteristics similar to entities which experienced serious problems.
Results from the 2 models are essentially identical based on 1 year prior
data. Model Z1 performed better for years 2 and 3; therefore, only the results
of that model are discussed. Of the 58 firms in the combined 2 samples, 7 are
misclassified, yielding an overall accuracy of 88%. The Type I error (that of
classifying a serious-problem firm as a continuing entity) was 13% (3 out of
23 misclassified) and the Type II error (that of misclassifying a continuing
entity) was slightly lower at 11.4% (4 of 35). These results are impressive
since they indicate that published financial data in Brazil, when correctly
interpreted and rigorously analyzed, does indeed possess important
information content.
Due to the potential upward bias involved in original sample classification
results, further tests of the models were performed with several types of
holdout or validation samples. The accuracy of the S-P sample is unchanged
184 E.I. Altman. Success of business failure models

after applying the Lachenbrunch test. Several replication tests also showed
high accuracy levels. Finally, the accuracy of the model is examined as the
data become more remote from the serious problem date. The S-P sample
results, as expected, show a drop in the accuracy of the models. We utilized
the weights from the model constructed with year 1 data and inserted the
variable measures for years 2 and 3 prior to the S-P date. Year 2 data
provided lower accuracy of 84.2% (16 of 19 correct). Year 3 data provided
accuracy of 77.8% (14 of 18 correct) classifications. Therefore, in only 4 cases
were errors observed in classification based on data from 3 (or more in some
cases) years prior to the S-P date.

4.4. Implications of results for Brazil

The implications and applications of models designed for assessing the
potential for serious financial problems in firms are many. This is especially
true in a developing country, where an epidemic of business failures could
have drastic effects on the strength of the private sector and on the economy
as a whole. Most observers of the Brazilian situation would agree on the
merit of preserving an equilibrium among private enterprises, state-owned
firms, and multinationals. Such equilibrium would be jeopardized if the
domestic private sector were weakened by an escalation of liquidations. If a
model, such as the one suggested, is used to identify potential problems, then
in many cases preventive or rehabilitative action can be taken. This should
involve a conscious internal effort, by the firms themselves, to prevent critical
situations as soon as a potential problem is detected. Besides internal efforts,
a program of financial and managerial assistance - more than likely from
official external sources - is a potential outcome.
Many economists, including this writer, have argued that significant
government assistance for the private sector is an unwise policy except
where the system itself is jeopardized. One can rationalize government
agencies’ attempts to stabilize those industries where a significant public
presence or national security is involved, e.g., commercial and savings banks
or a steel industry. In developing countries, the distinction between high
public interest sectors and the fragile private sector is more difficult to maxe
and limited early assistance is advocated.

4.5. Applying models in developing v. developed economies

We are fully aware of the problems that the wholesome and unqualified
use of failure models could cause. Obviously, if a firm demonstrates a profile
similar to past serious-problem entities and the model is used by all or most
of those organizations providing credit to firms, the model’s ‘prediction’ will
become a self-fulfilling prophecy. While some firms ‘deserve’ to cease their
 EL Altman, Success of business failure models 185

operations due to inefficiency, we advocate the potential of such

prognostications to help rebuild and restructure a temporarily faltering
operation. A recent study in the United States documents the efforts on the
part of a chief executive of an American Stock Exchange listed concern and
how he managed the firm back to financial health [Altman and LaFleur
(1981)]. In any event, the model should not be construed as either infallable
(as the results show clearly) or as a type of ‘black-box’ solution to problems.

5. Australia
Australia has certain unique characteristics with huge development
potential (like Brazil) but with an already established industrial base. While
the influence of multinational firms is quite important, the local corporate
structure is large enough to support a fairly substantial capital market. The
various types of financial and accounting organizations present have grown
in parallel with the corporate register. One of the characteristics of a
dynamic capitalistic economy is the flow of firms entering and exiting the
system. Despite strong growth characteristics, Australia also has experienced
a relatively high business failure rate which we estimate to be perhaps 2 or 3
times higher than that of the United States. In actuality, the Australian
liquidation figures are inflated due to the inclusion of voluntary ‘winding-ups’
on the part of firms which are not the result of financial distress. We estimate
[Altman and Izan (19Sl)] that the ‘true’ failure figures in 1978-1980 were in
the vicinity of 45-60 per 10,000 registered firms. In 1978 there were 57
voluntary liquidations, most of them not associated with distress. In any
event, the annual failure rate is less than 1% of the population.

5.1. Research design - the Castagna and Matolcsy model

The active financial environment in Australia is a motivation for rigorous

individual firm analysis. A series of studies by Castagna and Matolcsy
(C&M), culminating in their published work [C&M (1981)], have analyzed
the failure prediction potential in Australia and have concluded that there is
a strong potential for models, like those developed in the U.S., to assist
analysts and managers.
One of the dill&It requirements for failure analysis found in just about
every country in the world outside the U.S. is to assemble a data base of
failed companies large enough to perform a reliable discriminant analysis
model. Despite a relatively large number of liquidations, Australian data on
failed firms is quite restricted. C&M were able to assemble a sample of only
21 industrial companies (the number of firms would have been much larger if
mining companies were included). The failure dates spanned the period 1963-
1977, with the date determined by when a liquidator or receiver was
186 E.I. Altman, Success of business failure models

appointed. An alternative criterion date might have been the time of delisting
from the stock exchange or the liquidation/receiver date, whichever comes
first. For every failed company in the sample, there is a randomly selected
surviving quoted industrial firm from the same period. Industries represented
include retailers, manufacturers, builders, and service firms.

5.2. Empirical results

Prior studies by C&M reduced the number of potential discriminating
variables to 10 which were then analyzed in a linear and quadratic
discriminant structure. The authors also attempted to test their results for
various a priori group membership probabilities. The results suggest that it is
difficult to identify a unique model to predict corporate failures and that
some specification of user preferences is desirable. Still, they do indicate a 10
variable linear and 5 variable quadratic classification model which are
specified in detail in C&M (1981).
As noted, the results of C&M’s work are not definitive. For example, if
one is concerned with minimizing the misclassification of failed companies,
then the linear model using equal priors outperforms all other models tried.
This model also had the best overall results, except in the fourth year prior
to failure. However, the linear model does not perform better than other
models in the classification of surviving companies. A step-wise procedure
indicated that a 5 variable model did not perform as well as the models
based on the 10 ratio set in the overall classification tests. All of their
comparisons are based on the Lachenbruch validation tests.
The C&M study does not address prediction accuracy per se. All of the
tests are on the original sample of 21 firms. In order for the tests to be
predictive in nature, their model(s) should be applied to subsequent firm
performance in Australia. The authors do note that they expect to monitor
their findings on samples of continuing companies listed on the Australian
Stock Exchange.

Recent models from Australia. Two independent studies, one by Izan and
the other by Lincoln, provide even more recent refinements to the
bankruptcy classification problem in Australia and will be discussed in this
Special Issue.

6. England
The British financial environment is probably the most similar one to the
United States system in terms of the amount and quality of financial
information available to analysts of individual corporate entities. The
number and rate of business failures in the U.K. is perhaps the highest in the
 E.I. Altman, Success of business failure models 187

world, especially in the late 1970s and early 1980s when relatively tight
monetary policy and recessionary periods wracked the economy. Table 4
shows that the failure rate in England over the last decade has averaged over
twice that of the U.S. and, based on earlier comparisons, is significantly
higher than Japan, West Germany and Australia. As such, it is
understandable that attempts have been made in the U.K. to analyze the
information content of reported data for assessing firm performance and
failure potential.

Table 4
Business failure rates, U.S. and U.K.,
1971-1980 (percent of firms registered
or covered).

U.S.(%) U.K. (%)

1971 0.42 0.69
1972 0.38 0.59
1973 0.36 0.44
1974 0.38 0.64
1975 0.43 0.94
1976 0.35 0.99
1977 0.28 0.96
1978 0.24 0.77
1979 0.28 0.65
1980 0.42 0.95

“Based on Dun & Bradstreet’s

business failure rate; the percentage of
firms followed by D&B (converted
from the number per 10,000 fnms to
a percent); from Dun & Bradstreet
(various issues).
bCompany liquidations in the U.K.
as a percentage of registered
companies; from Annual Abstract of
Statistics, London, 1981, and the
Annual Report of the NIDC, 1979.

Taffler has written a comprehensive survey of failure classification models

from the U.K. as well as an update on his numerous works. That paper is
included in this Special Issue.

7. Ireland
In Ireland, Cahill (1981) presents some exploratory work on a small
sample of 11 bankrupt, listed companies covering the period 197&1980.
Three primary issues are explored: (1) identification of those ratios which
showed a significant deterioration as failure approaches, (2) whether the
auditors’ reports expressed any reservations or uncertainty about the
188 E.I. Altman, Success of business failure models

continuance of the firms as a going concern, and (3) whether there were any
other unique aspects of the failed companies’ condition.
Cahill’s analysis revealed a number of ratios indicating clear distress
signals 1 year prior to failure. These ratios compared unfavorably with
aggregate norms and ratios for the comparable industrial sector. Although
several measures continued to show differences in earlier years, the signals
were less clear in year 2 prior and it was difficult to detect strong signals
from ratios prior to year 2.
Only 1 of the 11 auditors’ reports was qualified on the basis of going
concern. A number (5) of other less serious qualifications were present in the
auditor reports. Cahill speculates that the low frequency of auditor
qualifications based on a going concern basis was due to auditor reluctance
and accounting convention in Ireland as well as being part of a ‘small
society’. We observed similar circumstances in Australia. Still, according to
Cahill, since deterioration was quite apparent, those close to the situation
should have been aware of the seriousness and earlier remedial action taken
or qualification given.
Unsuccessful merger activity and significant investment and asset
expansion financed by debt were the major causes of Irish failures. Several of
the firms continued to pay dividends right up to the year prior to failure. On
the other hand, only one company actually made payments to unsecured
creditors after insolvency, indicating that asset value had deteriorated beyond
repair and only then was failure declared.

8. Canada

Canada, like Australia, is a relatively small country in terms of business

population, yet it too is concerned with the performance assessment of
individual entities. The economy is very much tied to the fortunes of the
United States and its financial reporting standards are often derived from the
same accounting principles. Like so many environments that we have looked
at earlier, the key constraint in Canada is the availability of a large and
reliable data base of failed companies. This requires both a sufficient number
of failures and publicly available data on those firms. Both attributes do exist
in Canada but only just barely.
Knight (1979) analyzed the records of ‘a large number of small business
failures as well as conducting interviews with the key persons involved. The
author contends that his study supplies information ‘to answer the question,
why do small businesses fail in Canada and also generates certain guidelines
as to how the failure rate in Canada may be decreased from its recent
increasing level’. Not surprisingly, Knight finds that a firm usually fails early
in its life (50% of all failed firms do so within 4 years and 70% within 6) and
that some type of managerial incompetence accounts for almost all failures.
 E.I. Altman, Success of business failure models 189

Knight also attempted to classify failure using a discriminant analysis

model. He amassed a fairly large sample of 72 failed small firms with average
sales and assets of about $ 100,000. A five variable discriminant function
realized disappointing results, however. Only 64% of the original sample of
36 failed and 36 non-failed firms and 54% of the test sample of a like number
of firms were correctly classified. He concluded that the discriminant analysis
procedure was not successful. Knight did combine firms in many different
industries including manufacturing, service, retail, and construction, and this
will contribute to estimation problems especially if the data is not adjusted
to take into consideration industry differences and/or accounting differences,
e.g., lease capitalization. We discussed this industry effect at length earlier in
the Australian situation.

8.1. Altman and Lavallee’s work

The results of Altman and Lavallee (A&L) (1981) were more accurate
when manufacturing and retailing lirms were combined but they do not
advocate a single model for both sectors. Indeed, the holdout tests of this
study indicate that non-manufacturers cannot be confidently measured when
the model contains variables which are industry sensitive.
The A&L study was based on a sample of 54 publicly traded firms, half
failed and half continuing entities. The failures took place over the decade
1970-1979 and the average tangible asset size of these 27 failures was $12.6
million at the one statement date prior to failure (average lag was 16
months). Manufacturers and retailer-wholesalers were combined although
the data did not enable them to adjust assets and liabilities for lease
capitalization. The continuing firms were stratified by industry, size and data
period and had average assets of $ 15.6 million. One can observe, therefore,
that the Canadian model for the 1970s decade consisted of firms with asset
sizes similar to the previously reported early U.S. models [e.g., Altman
(1968)] constructed from the 1950s and 1960s data period.
A&L examined just 11 ratios and their resulting model contained 5 based
on a forward step-wise selection procedure. The model for Canada (23 is

Z,= -1.626+0.234X, -0.531X,+1.002X, +0.972X,+0.612X,,

where
2, = Canadian Z-score,
X1 = sales/total assets,
X, = total debt/total assets,
X, = current assets/current liabilities,
X, =net profits after tax/total debt,
X, =rate of growth of equity-rate of asset growth.
190 E.I. Altman, Success of business failure models

8.2. Classification results

The overall classification accuracy of the 2, model on the original 54 firm
sample was 83.3x, which is quite high, although not as impressive as that
reported for some of the other economic environments. Practically speaking,
classification criteria is based on a zero cutoff-score with positive scores
indicating a non-failed classification and negative scores a failed assignment.
Reliability, or holdout tests, included the Lachenbruch (1967) test
replications of the original sample broken into randomly chosen classification
and test samples, and testing the model on prior years data, e.g., years 2-4
before failure. The Lachenbruch and replication holdout results showed
accuracies very similar to the original sample results and the prior year
accuracies were 73% (year 2), 53% (year 3), and only 50% (year 4). Therefore,
the model appears reasonably accurate for up to 2 statements prior to failure
but not accurate for earlier periods. These findings are quite similar to those
of Altman’s (1968) model and we can suggest that the similarities in
accuracies are partially related to the similarities of the data quality and the
somewhat diverse industries represented in the sample.
A&L also simulated the results for various assumptions of prior
probabilities of group membership and costs of error. Their findings were
that Type I errors could be reduced, even eliminated, but that the resulting
Type II error was unacceptably high and vice versa for eliminating the Type
II error. The 2, model’s results were also compared to a naive classification
strategy of assigning all observations to the non-bankrupt category or
assuming that the resulting errors would be realized in proportion to the
actual experience of bankrupts and non-bankrupts [proportional chance
model, see Joy and Tollefson (1975)]. They concluded that, in every case, the
Canadian 2, model was more efficient, i.e., had a lower expected cost than a
naive model.
Finally, A&L observe that the industry afllliations of the misclassified
firms were predominantly retailers amongst the failed group and
manufacturers amongst the non-failed. It appeared that one of the variables,
sales/assets (X,), was particularly sensitive to industry effects with the
misclassified failed retailers all having high asset turnovers and the
misclassified manufacturers all with low turnovers.

8.3. Implications
A&L attempted to re-estimate the model without the sales/asset variable
but the results actually were worse. One. can conclude that the Canadian
investigations are at an early stage and follow-up work is needed in
subdividing a larger sample into manufacturers and retailer-wholesalers
and/or improving the information on critical industry differences, such as
lease usage and capitalization. Only additional time will permit analysts to
 E.I. Altman, Success of business failure models 191

construct models with sufficiently large samples or to witness an

improvement in the quality of reported data. We were aware of a move with
the Canadian government to set up an early warning system to identify
potentially large, publicly-traded, firm crisis situations, e.g., Massey-
Ferguson. Authorities are currently considering available models, e.g.,
Altman (1968) and A&L (1981), as an alternative to building their own
model.

9. The Netherlands
Another relatively small country which has received attention from
financial researchers of distress prediction is The Netherlands. Several recent
Dutch studies have attempted to apply classification techniques to failure
early warning systems. The main problem that Abrahams and van
Frederikslust (1976), Bilderbeek (1977,1979) and van Frederikslust (1978)
encountered was small sample size of bankrupt firms. Still, their works were
carefully performed and added evidence that the financial ratios can indeed
convey a great deal of information.

9.1. Bilderbeek

Bilderbeek (1977) analyzed a sample of 38 firms which went bankrupt from

1950-1974 and 59 ongoing companies, but found that only 85 firms had
sufficient data for analysis. Bilderbeek analyzed 20 ratios within a step-wise
discriminant framework and arrived at a 5 variable model of the form

ZNB= Z-score (Netherlands, Bilderbeek),

X1 =retained earnings/total assets,
X2 = added value/total assets,
X, = accounts payable/sales,
X4 = sales/total assets,
X5 = net profit/equity.

Two of the 5 signs (coefficients), X, and X,, are positive and contrary to
expectations since, for this model, negative scores indicate a healthy situation
and positive scores indicate a failure classification. His model was based on
observations over 5 reporting periods prior to failure and is not based on 1
year intervals. His results were only mildly impressive with accuracies
ranging from 70-80% for 1 year prior and remaining surprisingly stable over
a 5 year period prior to failure. He explains in his book (1979) that the
192 E.I. Altman, Success of business failure models

reason for the stability is that there are no liquidity variables and also the
stable role of his value added measure. Subsequent tests of Bilderbeek’s
model have been quite accurate (8O’k over 5 years). Apparently, several
institutions are now using his model for practical purposes.

9.2. Van Frederikslust

Van Frederikslust’s (1978) model included tests on a sample of 20 failed
and a matched non-failed sample of observations for the period 19541974.
All firms were quoted on the Netherlands Stock Exchange. In addition to the
now traditional research structure, i.e., linear discriminant, single year ratio,
equal a priori probability of group membership assumptions, the author
performed several other tests. Those included (1) looking at the development
of ratios over time (temporal model) as well as analyzing ratio levels, (2)
varying the a priori assumption of group membership likelihood to conform
with a specific user of the model (e.g., lending-officer), and (3) varying the
expected costs to the models taking into consideration the specific user’s
utility for losses.
Van Frederikslust attempted to provide a theoretical discussion for his
choice of variables. He concludes that traditional measures of firm
performance (i.e., liquidity, profitability, solvency, and variability of several of
these categories) are the correct indicators to look at. Industry affiliation and
general economic variables are also thought to be important but are not
included in his model. In fact, the primary model only contained 2 variables
representing liquidity and profitability.
Van Frederikslust’s primary model analyzed level of ratios. His definition
of failure included many different types of juridical mechanisms but
essentially involved the failure to pay fixed obligations. His sample included
textile, metal processing, machinery, construction, retailing, and
miscellaneous firms. The non-failed group (20) were randomly selected from
the same industries, size (assets), and time periods as the failed group. His
first model was

Z,, = 0.5293 +0.4488X, +0.2863X,,

where
Z,, = Z-score (Netherlands, Frederikslust),
X, = liquidity ratio (external coverage),
X, =profitability ratio (rate of return on equity).

The author distinguishes between the internal coverage ratio (cash balance
+resources earned in the period/short term debt) and the external coverage
ratio [short term debt in period t plus available short term debt (Q/short
 EL Altman, Success of business
failure models 193

term debt (t- l)]. The external coverage measures what can be expected
from the renewal of debt and additional debt. Van Frederikslust (1978, p. 35):
‘Failure at moment (t) is completely determined by the values of internal and
external coverage at that moment.’ van Frederikslust uses only the external
coverage measure in his ‘simple’ model.
Separate models are developed for each year, like Deakin (1972). The
arguments for this are that a separate model is necessary to assess failure
probabilities for different time periods and that the distributions of ratios
vary over time. While we do not necessarily agree that separate models are
desirable - indeed they could be confusing - the discussion on timing of
failure prediction is a useful one. The classification program utilized was
actually a 0,l multiple regression structure and not the discriminant analysis
model used in most other studies. Fisher (1936) has shown that the
coefficients of these structures are proportional when dealing with a 2 group
model.
The results for the 1 period model indicate that the estimated chances of
misclassification into the 2 groups are 5% for the failed group and 10% for
the non-failed group. The expected accuracy falls as time prior to failure
increases. For example, the error rates are 15% and 20x, respectively, for 2
years prior.
A revised model, analyzing the development of ratios over time, yielded an
equation which utilized the liquidity ratio in the latest year before failure, the
profitability ratio 2 years prior, the coefficient of variation of the liquidity
ratio over a 7 year period, and the prediction error of the pro~tability ratio
in the latest year before failure. Again, separate models were developed for
each year prior to failure. Using Lachenbruch’s procedure for estimating
error rates, the results were quite similar to the first set of equations based
on the 2 variable, ‘levels’ ratios. Accuracies for earlier years did show slight
improvements.

Iin France, Altman et al. (1974) first attempted to apply credit scoring
techniques to problem firms, many of which filed for bankruptcy (faillite).
Working with a sample of textile firms and data provided by Banque de
France, this study applied principal component analysis to a large number of
financial indicators and proceeded to utilize the most important ones in a
linear discrim~nant model. Their results were at best mediocre on test
samples and, whiie the model did provide insights into that troublesome
sector, it was not implemented on a practical basis.
A more recent study by Bontemps (1981), using a large sample of
industrial companies and data from the CentraIe de Bilans of Credit
National (supplier of long term debt capital to’French firms), achieved high
194 E.I. Altman, Success of business failure models

accuracy on original and holdout tests. His results are quite interesting in
that as little as 3 variables were found to be useful indicators. Bontemps
combined the univariate technique developed by Beaver (1966) with
arbitrary, qualitative weightings of the 3 most effective measures to correctly
classify as much as 87% of his holdout sample of 34 failed and 34 non-failed
firms. The original function was built based on a matched (by industry, size,
and year) sample of 50 failed and non-failed entities over the period 1974-
1979. At this time, the results are being evaluated and, of course, they are
confidential.
Collongues (1977), Mader (1975,1979) and Bontemps (1981) also have
attempted to combine financial ratios with data from failed and non-failed
French firms. Mader’s studies were descriptive of firms in difficulty and the
utility of ratios as risk measures. Collongues did utilize discriminant analysis
in his analysis of small and medium size firms with some success.
The application of statistical credit scoring techniques in the French
environment appears to be problematic but the potential remains. One
problem usually is the quality of data and the representativeness of it. But
this is a problem in all countries and is not unique to France. It would
appear that the new economic plan, being put forth by the Socialist
government in France in 1981-1982, could very well utilize a well
constructed and reliable model. The government has gone on record not to
artificially keep alive hopelessly insolvent firms but to try and assist those
ailing firms prior to total collapse. An accurate performance predictor model
could very well help in this endeavor. In this Special Issue, Micha explores
the attempts at Banque de France.

11. International comparison of ratios

In putting together this paper, an interesting issue arose as to the

similarities between ratios of failed and non-failed firms across national
boundaries. Since many of the studies reviewed earlier utilized the~same, or
similar, variable profiles it was possible to examine average ratios in different
countries.
The comparative average ratio results for 4 different countries (5 samples
of firms since we examined 2 separate studies from the United States) are
listed in table 5. In addition to the U.S., these countries include Canada,
Brazil, and Australia. We do not have any a priori economic hypothesis as to
expected ratio similarities across countries but, due to financial reporting
standards unique to local conditions and different political-economic
interrelationships, we do not expect a great deal of similarity. Indeed, we do
find large differences in average ratios for our failed groups and somewhat
smaller differences across non-failed groups. Since we did not have the actual
distributions of firm ratios foi most of our samples, it was not possible to do
any significance tests across samples.
 E.I. Altman. Success of business failure models 195

Table 5
International comparisons of average group ratios.”

U.S. Australia Brazil Canada

U.S. Altman Castagna & A&man Altman & Japan
Altman et al. Matolcsy et al. Lavallee Ko
Ratio name 1968 1911 1981 1979 1981 1981

Failed groups
Working capital/
total assets -0.061 0.150 0.062 -0.120 0.100 -0.181
Retained earnings/
total assets -0.626 - 0.406 -0.038 0.010 n.a. -0.163
EBIT/total assets -0.318 -0.005 0.002 0.050 -0.120 - 0.077
Market value equity/
total liabilities 0.401 0.611 0.800 0.350 n.a. 0.533
Sales/total assets 1.500 1.310 1.200 0.880 1.480 1.052
Average Z-score - 0.266 1.271 1.707 1.124 n.a. 0.667

Non-failed groups
Working capital/
total assets 0.414 0.309 0.187 0.230 0.300 0.107
Retained earnings/
total assets 0.355 0.294 0.22 0.240 n.a. 0.154
EBIT/total assets 0.153 0.112 0.086 0.160 0.040 0.063
Market value equity/
total liabilities 2.477 1.845 3.110 1.140 n.a. 0.878
Sales/total assets 1.900 1.620 n.a. 1.230 2.310 0.988
Average Z-score 4.883 3.878 4.003 3.053 n.a. 2.070

“Sources: Various articles, see reference section.

With the exception of my early study [Altman (1968)], we observed that

the data is derived from essentially similar time periods for the other 4
samples. Most involved data from the decade 1970-1979. Therefore, we
cannot explain ratio mean differences based on temporal grounds. We
observe fairly large cross-country differentials in all ratios for the failed
groups and, with the exception of the asset turnover, all of the failed groups
ratio averages were ‘better’ than those in the earlier 1968 study. Cross
ranking of averages for individual ratios do not reveal consistent country
rankings for either the failed or the non-failed groupings. We conclude,
therefore, that ratio averages are not ‘country sensitive’ in a consistent
manner.
We also can compare the average ratios for two U.S. samples (1968
and 1977). In every ratio case for the non-failed group, the later figures (1977)
show a deterioration in the average ratio. Two reasons could account for
this. One, there has been a general deterioration since the 1960s in the
financial performance of U.S. corporations, and financial risk has increased.
Second, the 1977 data is adjusted for capitalization of leases, thereby
196 E.I. Altman, Success of business failure models

increasing the asset base. With a larger denominator in each of the ratios, we
observe lower positive ratios and ‘higher’ negative ratios. This is confirmed by
the failed group averages, which shows that, for negative ratios, the averages
are higher in the 1977 period. The only exception to the lease capitalization
and financial deterioration theses is the somewhat higher average in 1977 for
the market value of equity/total liabilities ratio for failed firms. We find it
difficult to explain this, especially since the non-failed group showed the
opposite ranking.
We have also calculated, in table 5, average Z-scores for the bankrupt and
non-bankrupt groups in 4 countries (for 5 samples if we include the 2 from
the U.S.). It appears that the original Z-score (1968) sample had the greatest
dispersion with the highest average non-bankrupt and the lowest average
bankrupt firm score. The distribution is tightest for Ko’s Japanese sample.
Interestingly enough, the lowest non-bankrupt group’s average score
(i.e., Japan’s 2.07) is greater than the highest bankrupt group’s rating
(Australia’s 1.50).
We were also a bit surprised to find that the average bankrupt firm for the
Zeta (1977) sample was considerably higher than the earlier 1968 sample
(1.27 v. -0.27). A good deal of this difference is caused by our adjustments
to the data in the later sample (primarily the capitalization of leases). The
larger asset base tends to increase the negative ratios, particularly X,
(retained earnings/total assets) and X, (EBIT/total assets). This is consistent
with the lower average positive score for the non-bankrupt group in the Zeta
sample v. the earlier sample.
It is particularly interesting to compare the average ratios from the 2 U.S.
samples of non-failing firms [columns (1) and (2) in table 51 with the
Japanese sample. As we noted earlier, Japanese firms will probably have
lower Z-scores due to lower retained earnings, profits, and higher leverage as
well as higher asset totals per dollar (yen) of profits and sales. In fact, every
ratio for the 2 U.S. samples is greater than its Japanese counterpart. Note,
however, that while this is true for the non-failed sample, which we expect is
fairly representative for the population of firms, the distinction is not as clear
for the failed sample.

12. A few conclusions

In this paper we review and compare a relatively large number of
empirical failure classification models from 10 countries. Much of the
material is derived from little known or unknown sources and as such we
hope that the study will stimulate a greater trans-national discussion. Indeed,
as financial institutions and government agencies in various countries, e.g.,
Canada, U.S., Brazil, France, and England, wrestle with the problem of large
firm failures in the future, the knowledge that prior work has been done with
 E.I. Altman, Success of business failure models 197

respect to early warning models may help obviate the consequences or

reduce the number of these failures.
I expect the quality and reliability of models constructed in many of the
reviewed countries to improve as (1) the quality of information on companies
is expanded and refined, (2) the number of business failures increases thereby
providing more data points for empirical analysis, and (3) as researchers and
practitioners become more aware of the problems and potential of such
models. Where sufficient data does not exist for specific sector models, e.g.,
manufacturing, retailing, and service firms, the application of industry
relative measures can perhaps provide a satisfactory framework for
meaningful analysis. Of course, this requires that government or private
agencies build reliable industry data bases for comparison purposes.

References
Abrahams, A. and R.A.I. van Frederikslust, 1976, Discriminant analysis and the prediction of
corporate failure, in: R. Brealey and G. Rankine, eds., European finance association 1975
proceedings (North-Holland, Amsterdam).
Altman, E., 1968, Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy, Journal of Finance, Sept.
Altman, E., 1980, Commercial bank lending: Process, credit scoring and costs of errors in
lending, Journal of Financial and Quantitative Analysis, Nov.
Altman, E., 1983, Corporate financial distress (Wiley, New York).
Altman, E. and H.Y. Izan, 1981, Research in progress (Australian Graduate School of
Management, Sydney).
Altman, E. and J. LaFleur, 1981, Managing a return to financial health, Journal of Business
Strategy, Summer.
Altman, E. and M. Lavallee, 1981, Business failure classification in Canada, Journal of Business
Administration, Summer.
Altman, E. and T. McGough, 1974, Evaluation of a company as a going concern, Journal of
Accountancy, Dec.
Altman, E., T. Baidya and L.M. Riberio-Dias, 1979, Assessing potential financial problems of
firms in Brazil, Journal of International Business Studies, Fall.
Altman, E., R. Haldeman and P. Narayanan, 1977, ZETA analysis: A new model to identify
bankruptcy risk of corporations, Journal of Banking and Finance, June.
Altman, E., M. Margaine, M. Schlosser and P. Vernimmen, 1974, Statistical credit analysis in the
textile industry: A French experience, Journal of Financial and Quantitative Analysis, March.
Beaver, W., 1966, Financial ratios as predictors of failure, Empirical Research in Accounting:
Selected Studies, Supplement to Journal of Accounting Research 4.
Beerman, K., 1976, Possible ways to predict capital losses with annual financial statements
(Diisseldorf).
Bilderbeek, J., 1977, FinanciEle ratio analyse (Stenfert-Kroese, Leiden).
Bilderbeek, J., 1979, An empirical study of the predictive ability’ of financial ratios in the
Netherlands, Zeitschrift Fiir Betriebswirtschaft. Mav (in English).
Bonds, W., 1978, Business failures and the banks, ‘Euromoney, july:
Bontemps, P.O., 1981, La notation du risque de credit (Credit scoring for risk) (Credit National,
Paris).
Castagna, A.D. and Z.P. Matolcsy, 1977, An examination of company deaths: A comparative
study of the financial profiles of acquired and failed companies, Research paper no. 1 (Centre
for Securities Industries Studies, Kuring-gai College of Advanced Education, Sydney).
Castagna, A.D. and Z.P. Matolcsy, 1981, The prediction of corporate failure: Testing the
Australian experience, Australian Journal of Management, June.
Collongues, Yves, 1977, Ratios financiers et p&vision des faillites des petites et moyennes
entreprises (Financial ratios and forecasting failures of small and medium size enterprises),
Revue Banque, no. 356.
198 E.I. Altman, Success of business failure models

Deakin, E., 1972, A discriminant analysis of predictors of business failure, Journal of

Accounting Research 10, no. 1.
Dun & Bradstreet, Various issues, Failure record (New York).
Earl, M.J. and David Marais, 1982, Predicting corporate failure in the U.K. using discriminant
analysis, Accounting and Business Research.
Fischer, J., 1981, Forecasting company failure by the use of non-numerical data, Paper
presented at the EIASM Workshop on Bank Planning Models, Brussels, April 6-7.
Fisher, R.A., 1936, The use of multiple measurements in taxonomic problems, Annuals of
Eugenics, Sept.
Gebhardt, G., 1980, Insolvency prediction based on annual financial statements according to
the company law - An assessment of the reform of annual statements by the law of 1965
from the view of external addresses, in: H. Besters et al., eds., Bochurner Beitrage zur
Unternehmungs und Unternehmens-forschung, Vol. 22 (Wiesbaden).
Joy, O.M. and J. Tollefson, 1975, On the financial applications of discriminant analysis, Journal
of Financial and Quantitative Analysis 10, Dec.
Kanitz, S., 1974, Como prever y falencia de empresas, EXAME, Dec.
Knight, R.M., 1979, The determinants of failure in Canadian firms, ASA Meetings of Canada,
Saskatoon, May 28-30, Working paper (University of Western Ontario, London) May.
Ko, C.J., 1982, A delineation of corporate appraisal models and classification of bankruptcy
firms in Japan, Thesis (New York University, New York).
Lachenbruch, P., 1967, An almost unbiased method of obtaining confidence intervals for the
probability of misclassification m discriminant analysis, Biometrics, Dec.
Mader, F., 1975, Les ratios et l’analyse du risque (Ratios and the analysis of risk), Analyse
Financiere, 28me trimestre.
Mader, F., 1979, Un echantillon d’entreprises en difficult& (A sample of enterprises in
difficulty), Journee des Centrales de Bilan.
Marais, D.A.J., 1979, A method of quantifying companies relative financial strength, Working
paper no. 4 (Bank of England, London).
Nakki, F., 1976, FINPLAN - A model for credit rating in Finland (Kansallis-Osake Pakki
Bank, Helsinki).
Nikkei-Business, 1978, Z-score - There are 90 corporations whose score was below 1.0, June 5.
Sorter, G. and G. Benston, 1960, Appraising the defensive position of a firm: The interval
measure, Accounting Review, Oct.
Taffler, R., 1976, Finding those firms in danger, Accountancy Age, July 16.
Taffler, R., 1981, Forecasting company failure in the U.K., Working paper no. 23 (City
University Business School, London).
Tamer, R. and H. Tisshaw, 1977, Going, going, gone - Four factors which predict
accountancy, March.
Tnffler, R. and Houston, 1980, How to identify failing companies before it is too late,
Professional Administration, April.
Takahashi, K. and E. Altman, 1981, Conference on the State of the Art in Bankruptcy
Classification Models, June 25 (Keio Graduate School of Business, Yokohama).
Takahashi. K.. Kurokawa and K. Watese, 1979, Predicting cornorate bankruntcv through
financial statements (Society of Management Science of Keio Umversity) Nov. _ _ -
Tamari, 1966, Financial ratios as a means of forecasting bankruptcy, Management International
Review 4.
van Frederikslust, R.A.I., 1978, Predictability of corporate failure (Martinus Nijhoff Social
Science Division, Leiden/Boston).
von Stein, J.H., 1968, Insulvenzen Privater Banken (University of Miinchen, Mtinchen).
von Stein, J.H., 1981, Identifying endangered firms (Hohenheim University, Stuttgart-
Hohenheim).
Webb, L., 1980, ,Predicting Australian corporate failures, Chartered Accountant in Australia,
Sept.
Weibel, P.F., 1973, The value of criteria to judge credit worthiness in the lending of banks
(Bern/Stuttgart).
Weinrich, G., 1978, Prediction of credit worthiness, direction of credit operations by risk classes
(Wiesbaden).