Accounting Research Center, Booth School of Business, University of Chicago

Financial Ratios and the Probabilistic Prediction of Bankruptcy

Author(s): James A. Ohlson
Source: Journal of Accounting Research, Vol. 18, No. 1 (Spring, 1980), pp. 109-131
Published by: Wiley on behalf of Accounting Research Center, Booth School of Business,
University of Chicago
Stable URL: http://www.jstor.org/stable/2490395
Accessed: 09-07-2015 17:52 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Wiley and Accounting Research Center, Booth School of Business, University of Chicago are collaborating with
JSTOR to digitize, preserve and extend access to Journal of Accounting Research.

http://www.jstor.org

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

JournalofAccountingResearch
Vol. 18 No. 1 Spring1980
Printed in U.S.A.

Financial Ratios and the

Probabilistic Prediction
of Bankruptcy
JAMES

A. OHLSON*

1. Introduction
This paper presentssome empiricalresultsof a studypredicting
Therehave
corporatefailureas evidencedby the eventofbankruptcy.
of
in
this
field
of
a
number
studies
fair
been
research;themore
previous
contributions
are
Beaver
1968a;
1968b],Altman
[1966;
notablepublished
[1968; 1973],Altmanand Lorris[1976],Altmanand McGough[1974],
Altman,Haldeman,and Narayanan[1977],Deakin[1972],Libby[1975],
Blum [1974],Edmister[1972],Wilcox[1973],Moyer[1977],and Lev
papersby Whiteand Turnbull[1975a; 1975b]
[1971].Two unpublished
as
and a paperby Santomeroand Vinso[1977]are ofparticular
interest
be
the
first
studies
which
and
to
systematically
logically
theyappear
estimatesoffailure.The presentstudyis similarto
developprobabilistic
is one ofmaximum
likelihood
thelatterstudies,in thatthemethodology
oftheso-calledconditional
estimation
logitmodel.
The data setusedin thisstudyis fromtheseventies(1970-76).I know
of onlythreecorporatefailureresearchstudieswhichhave examined
data fromthisperiod.One is a limitedstudyby Altmanand McGough
[1974]in whichonlyfailedfirmsweredrawnfromthe period1970-73
error(misclassification
offailedfirms)
and onlyone typeofclassification
was analyzed.Moyer [1977] consideredthe period 1965-75,but the
firms
The
was unusuallysmall(twenty-seven
firms).
sampleofbankrupt
* Associate Professor,Universityof California,Berkeley. I gratefullyacknowledgethe
financialsupportfromthe Wells Fargo Bank. My thanksare also due to R. Wagnerand R.
Benin, who providedable and valuable assistance in the course of the project.G. Feltham,
R. Hamilton, V. Anderson,W. Beaver, and R. Holland supplied valuable commentson
earlierversionsof this paper. [Accepted forpublicationMarch 1979.]
109

Copyright?, Instituteof ProfessionalAccounting1980

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

110

JOURNAL

OF

ACCOUNTING

RESEARCH,

SPRING

1980

third study,by Altman, Haldeman, and Narayanan [1977], which "updates" the originalAltman [1968] study,basically considers data from
failed
the period 1969 to 1975. Their sample was based on fifty-three
firmsand about the same number of nonfailedfirms.In contrast,my
studyrelies on observationsfrom105 bankruptfirmsand 2,058nonbankrupt firms.Althoughthe otherthree studies differfromthe presentone
so far as methodologyand objectives are concerned,it is, nevertheless,
interestingand useful to compare theirresultswith those presentedin
this paper.
Another distinguishingfeature of the present study which I should
stress is that, contraryto almost all previousstudies,the data forfailed
firmswere not derivedfromMoody's Manual.' The data were obtained
instead from 10-K financial statementsas reported at the time. This
procedure has one importantadvantage: the reports indicate at what
point in time they were released to the public, and one can therefore
check whetherthe companyenteredbankruptcypriorto or afterthe date
of release. Previous studies have not explicitlyconsidered this timing
issue. Some studies,but by no means all, seem implicitlyto presumethat
a reportis available at the fiscalyear-enddate. The lattermay or may
not be appropriate,dependingon the purpose of the study.However,if
the purpose is one of investigatingpure forecastingrelationships,as is
the case in this study, then the latter procedure is inadequate. This
followsbecause it is possible that a companyfilesforbankruptcyat some
point in time afterthe fiscalyear date, but priorto releasingthe financial
statements.This is not a trivialproblem and neglectingthis possibility
may lead to "back-casting"formany of the failedfirms.
The major findingsof the study can be summarizedbriefly.First, it
was possible to identifyfourbasic factorsas beingstatisticallysignificant
in affectingthe probabilityoffailure(withinone year). These are: (i) the
size of the company; (ii) a measure(s) of the financialstructure;(iii) a
measure(s) of performance;(iv) a measure(s) of currentliquidity (the
evidence regardingthis factoris not as clear as compared to cases (i)(iii)). Second, previous studies appear to have overstatedthe predictive
(in the sense of forecasting)power of models developed and tested. The
point of concern is the one alluded to above, that is, if one employs
predictorsderivedfromstatementswhichwere released afterthe date of
bankruptcy,thenthe evidence indicatesthat it will be easier to "predict"
bankruptcy.However,even ifone allows forthisfactor,forthe sample of
firmsused in this study,the predictionerror-rateis largerin comparison
to the rate reportedin the originalAltman [1968] study as well as most
other studies using data drawn fromperiods priorto 1970. More important, the prediction error-rateis also larger than the one reported in
Altmanet al. [1977]. On the otherhand,the Altmanand McGough [1974]
' The onlyexceptionappears to be the Altmanand McGough [1974] study.Altmanet al.
[1977] do not describe how theyderivedtheirdata.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

BANKRUPTCY

iII

whichare
largererror-rates,
and Moyer [1977] studiesreportsignificantly
comparable to those foundin this study.I have not been able completely
to account forthis most significantdifferencein the error-ratesreported
here,in Altman and McGough [1974], and in Moyer [1977], as compared
to Altman et al. [1977]. (Any period dependence should after all be
relativelyminor.)
The model(s) are relativelysimple to apply and may be of use in
practical applications. The data requirementsare such that all of the
predictors are easily retrieved from the Compustat file. A potential
disadvantage is that the model does not utilize any markettransactions
(price) data of the firms.One may, of course, expect that the predictive
power of the model could be enhanced by incorporatingsuch data.'
However, one mightask a basic and possibly embarrassingquestion:
why forecastbankruptcy?This is a difficultquestion,and no answer or
justificationis givenhere.It could,perhaps,be arguedthat we are dealing
with a problemof "obvious" practicalinterest.This is questionable since
real-worldproblemsconcernthemselveswithchoices whichhave a richer
set of possible outcomes.No decision problemI can thinkofhas a payoff
space which is partitionednaturallyinto the binarystatus bankruptcy
versus nonbankruptcy.(Even in the case of a "simple" loan decision,the
payoffconfigurationis much more complex.) Existingempiricalstudies
reflectthis problem in that there is no consensus on what constitutes
"failure," with definitionsvarying significantlyand arbitrarilyacross
studies. In otherwords,the dichotomybankruptcyversusno bankruptcy
is, at the most, a very crude approximationof the payoffspace of some
hypotheticaldecision problem. It follows that it is essentially a futile
exerciseto tryto establish the relativedecision usefulnessof alternative
predictivesystems.Accordingly,I have not concernedmyselfwith how
bankruptcy(and/or failure)"ought" to be defined;I also have refrained
frommaking inferencesregardingthe relative usefulnessof alternative
models, ratios, and predictivesystems (e.g., univariateversus multivariate). Most of the analysis should simplybe viewed as descriptivestatistics-which may, to some extent, include estimated predictionerrorrates-and no "theories" of bankruptcyor usefulnessof financialratios
are tested. Even so, there are a large numberof difficultstatisticaland
methodologicalproblems which need to be discussed. Many important
problemspertainingto the developmentof data forbankruptfirmshave
gone mostlyunnoticedin the literature.

2. Some CommentsRegarding Methodologyand Data

Collection
The econometricmethodologyof conditionallogitanalysiswas chosen
to avoid some fairlywell known problems associated with Multivariate
I am currentlyundertakingworkin this direction.I should note furtherthat the uac of
Hence, it can also be viewed
price data implicitlyis anotherway ofusingmoreinformation.
as another way of indirectuse of accountingdata.
2

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

112

JAMES

A. OHLSON

DiscriminantAnalysis (MDA, forshort). The MDA approach has been

the most popular technique for bankruptcystudies using vectors of
predictors.Among some of the problems with these studies are:3 (i)
There are certainstatisticalrequirementsimposed on the distributional
propertiesof the predictors.For example, the variance-covariancematrices of the predictorsshould be the same forboth groups (failed and
nonfailedfirms);moreover,a requirementofnormallydistributedpredictors certainlymitigatesagainst the use of dummyindependentvariables.
A violationofthese conditions,it could perhapsbe argued,is unimportant
(or simplyirrelevant)if the only purpose of the model is to develop a
discriminatingdevice. Althoughthis may be a valid point,it is nevertheless clear that thisperspectivelimitsthe scope ofthe investigation.Under
many circumstances,it is of interest to go throughmore traditional
econometricanalysis and test variables for statisticalsignificance,etc.
(ii) The output of the application of an MDA model is a score whichhas
little intuitive interpretation,since it is basically an ordinal ranking
(discriminatory)device. For decision problemssuch that a misclassification structureis an inadequate descriptionof the payoffpartition,the
score is not directlyrelevant.4If,however,priorprobabilisticsof the two
groupsare specified,thenit is possible to deriveposteriorprobabilitiesof
failure.But, this Bayesian revisionprocess will be invalid or lead to poor
approximationsunless the assumptions of normality,etc. are satisfied.
(iii) There are also certainproblemsrelatedto the "matching"procedures
which have typicallybeen used in MDA. Failed and nonfailedfirmsare
matched accordingto criteriasuch as size and industry,and these tend
to be somewhatarbitrary.It is by no means obvious what is reallygained
or lost by different
matchingprocedures,includingno matchingat all. At
the very least, it would seem to be more fruitfulactually to include
variables as predictorsratherthan to use them formatchingpurposes.
The use of conditional logit analysis, on the other hand, essentially
avoids all of the problems discussed with respect to MDA. The fundamental estimation problem can be reduced simply to the following
statement: given that a firmbelongs to some prespecifiedpopulation,
what is the probabilitythat the firmfails withinsome prespecifiedtime
period? No assumptionshave to be made regardingpriorprobabilitiesof
bankruptcyand/or the distributionof predictors.These are the major
advantages. The statistical significanceof the differentpredictorsare
obtained fromasymptotic(large sample) theory.To be sure, as is the
case in any parametricanalysis, a model must be specified,so there is
always room for misspecificationof the basic probabilitymodel. (AlSee also Eisenbeis [1977] and Joyand Tollefson[1975] forextensivediscussions.
The payoffpartitionwill be inadequate if it is not feasible to definea utilityfunction
over the two types of classificationerrors.Any economic decision problemwould typically
require a richerstate partition.
3

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

BANKRUPTCY

113

it is beyondthe confines
thoughit is possible to test formisspecification,
of this paper to discuss and reporton the resultsof such tests.)
Regardless of the virtuesof probabilisticpredictionover MDA, there
are importantproblemswithrespectto data collectionofbankruptfirms
which deserve preliminarydiscussion.This matterwas alluded to in the
introduction.Realistic evaluation of a model's predictiverelationships
recuiresthat the predictorsare (would have been) available foruse prior
to the event of failure.Now, it is of course true that annual reportsare
not publicly available at the end of the fiscal year, since the financial
statementsmust be audited. Previous studies have not mentionedthis
problem,at least not explicitly.This is not surprisingsince most previous
studies have used Moody's Manual to derive the pertinentfinancial
ratios) and the Manual does not indicate at what point in time the data
were made available. Anotherreason is that not all studies have been
concernedwithstrictforecastingrelationships.That is, whetheraccountingstatementswere publiclyavailable or not had no directbearingupon
thesubject at hand. One such case is Beaver [1968a; 1968b],who studied
whetherfinancialratios will reflectimpendingfailure.The timingissue
ca-nbe expected to be serious forfirmswhichhave a large probabilityof
failurein the firstplace. Such firmsare in poor shape and the auditing
process could be particularlyproblematicand time-consuming.
Thus, it
is somewhat riskyto assume that financialreportswere available, say,
three monthsafterthe end of the fiscalyear. There are otherdisadvantages associated withMoody's Manual. The data are oftenhighlycondensed, and it is generallycomplicated,if not impossible,to reconstruct
actual balance sheets and income statements.5Again,firmswhich are in
poor shape are particularlydifficult,
since one can neverbe sure whether
some ofthemanypossiblespecial itemshave been givenspecial treatment
in Moody's tabulation.6Moreover,it should be noted that the comparative schedules over the different
years are ex-postreconstructions,
and
items fromprevious years may have been restatedand may differfrom
the amountsorigirialiv
reported.At a nontrivialcost,thisproblemcan be
circumventedifone uses several annual editionsofMoody's forthe same
firm.
Clearly,much can be gained by improvingthe data base. The evaluation of the predictive classificationpower of a model should be more
realistic,and, more importanthere, the same should apply forstandard
tests of statisticalsignificance.This is not to suggestthat it is important
to have "super accurate" data forpurposes of developing(as opposed to
evaluating) a discriminatorydevice. It mightwell be that the predictive
quality of any mode1 is reasonably robust across a variety of datagatheringand estimatingprocedures.
The conclusion is based oil a few"case studies."
The summaries of taxes (loss carry-forward,
in particular)and measures of operating
performanceappear to be the most difficult
itemsto deal with.
6

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

114

JAMES A. OHLSON

3. Collection of Financial StatementData

The collection
ofdataforbankrupt
firms
requiresa definition
offailure
ofthepopulationfromwhichfirmsare drawn.In this
and specification
is purelylegalistic.
study,thedefinition
The failedfirms
musthave filed
in the sense of ChapterX, ChapterXI, or some other
forbankruptcy
notification
indicating
bankruptcy
proceedings.7
The populationboundariesare restricted
by:(i) theperiodfrom1970to 1976;(ii) theequityof
the companyhad to have beentradedon somestockexchangeor overthe-counter
(OTC) market;(iii) the companymustbe classifiedas an
The firstcriterion
industrial.
was chosensimplybecauseit is the most
recentperiod,and the cutoff
pointof 1970was selectedas a matterof
The secondcriterion
practicality.
excludessmallor privately
held corporations.This is crucial,sinceotherwise
theuse ofCompustat
firmsas
a sourceof nonbankrupt
firmswouldbe precluded.The thirdcriterion
excludesutilities,
transportation
companies,and financialservicescompanies (banks,insurance,brokerage,
REITs, etc.). Companiesin these
industries
arestructurally
different,
havea different
bankruptcy
environdata are,in somecases,difficult
ment,and appropriate
to obtain.
The following
procedureswereused to generatea listoffailedfirms
theinclusioncriteria.(1) A primary
was
satisfying
listingoffailedfirms
tabulatedfromthe Wall StreetJournalIndex (W.S.J.I.).Typeanddate
ofbankruptcy
wererecorded.Ifthenameofthecompanyindicatedthat
the firmin questionwas a nonindustrial,
thenit was excluded.(2) A
secondarylistingof firmswas tabulatedby excludingall firmson the
primary
listingwhichhad notbeentradedon one ofthestockexchanges
(or OTC) duringthe three-year
periodpriorto the date ofbankruptcy.
Ifa companycouldbe tracedto one oftheexchanges,
thentheexchange
was recorded.This kindofinformation
was derivedfromvariousstock
guidesissued by Moody's and Standard and Poor's. Of course,as a
practicalmatter,it was assumedthata stockhad notbeentradedifno
evidencecould be foundto that effect.(3) Attemptswere made to
sources
augmentthesecondary
listingbyexamining
othermiscellaneous
of data. This led to some relativelyminoradditionsco the listingof
firms.
bankrupt
However,in tracingbankrupt
firms
itseemedto methat
veryfewfirmswere omittedfromthe W.S.J.I.,so long as the firms
satisfiedtheinclusioncriteria.
The nextphase was one of actuallycollectingfinancialdata forthe
The objectivewas to obtainthreeyearsofdata priorto
bankruptfirms.
the date of bankruptcy.
Each reporthad to includethe balancesheet,
incomestatement,
fundsstatement,
and theaccountants'
report.In case
thelast availableaccountants'
reportexplicitly
statedthatthecompany
had filedforbankruptcy,
thena fourth
reportwas collected.All reports
wereretrievedfromthe StanfordUniversity
BusinessSchool Library,
'See Altman [1971] fora discussionof the differencebetween differenttypes of bankruptcyproceedings.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL RATIOS AND BANKRUPTCY

115

TABLE
1
Bankrupt Firms: Year, Type ofBankruptcy,and Exchange Listing
Year

Type
Chapter X
.
Chapter XI
Other or unknown
Totals ...

.
.

1970

1971

1972

1973

1974

1975

1976

0
1
0
1

2
4
0
6

2
14
5
21

1
20
6
27

1
18
0
19

0
14
1
15

0
14
2
16

New York Stock Exchange

..
AmericanStock Exchange ...........
Other*
* Over-the-counter
orregional
market
exchange.

..
..........

Totals
6
85
14
105
.

8
43
54

whichhas an extensivemicrofilmfileof 10-K reports.The relevantparts

of the 10-K reports were photocopied and subsequently coded. Some
firmshad to.be deleted fromthe sample because no reportwhatsoever
was available, but these were few. Other firms,again very few, were
deleted because theywerecorporateshells and had no sales. On the other
hand, no firmwas deleted because of its young(exchange) age, and some
had only one set of reports.
firnms
In the process of codingitemsfromthe annual reports,I noted that all
but one firmwhich went bankruptin 1970,and some of the 1971 firms,
had no fundsstatementin theirannual report.This was not trueforfirms
whichfiledforbankruptcyin subsequent years. Similar observationsare
applicable forfirmson the Compustatfile,althoughomissionsweremuch
less frequent.The SEC did not requirea fundsstatementuntilthe early
seventies.' I decided that firmswithout a funds statement should be
deleted, since it would have been impossible otherwiseto use ratios
derived fromthe fundsstatement.
The finalsample was made up of 105bankruptfirms.Basic information
regardingyear, exchange,and type of bankruptcyis given in table 1. As
one would suspect, relativelyfew firmswere listed on the NYSE, compared to the othertwo categorizations.Furthermore,Chapter XI bankkruptcywas apparentlymuch more frequentthan Chapter X.
I noted that while eighteenof the 105 firms(17 percent)had accountants' reportswhich disclosed that the companyhad enteredbankruptcy,
the fiscal year-end was prior to the date of bankruptcy.These reports
were deleted and reportsfromthe previousfiscalyear were substituted.
As a consequence, the average lead time between the date of the fiscal
year of the last relevant reportand bankruptcyis quite long, approximatelythirteenmonths.Table 2 shows the entirefrequencydistribution.
This lead time is quite a bit longer,comparedto what has been reported
in previousstudies. A cursoryreviewof the data indicatedthat the time
lag between the fiscalyear-endand the date of the accountants' report
8

19.

Funds statementshave been requiredsince September 30, 1971; see APB opinion No,

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

116

JAMES A. OHLSON
Iq

LO~

LO~

ei

t
CI

oo

~~0 I~
I

C,-

O
C

c1LC

ei

Cl1
Cl1
o

t-

Co1
LO
I-1

.4

Ci2

CO

C i2

4 0

Econ

00

; n

Dz *

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

BANKRUPTCY

117

can be quite long. In fact,for the group of eighteen firms,the lag was
always longerin the year whichwas closer to bankruptcy.The following
example is reasonably representative.Hers Apparel Ind. filedforbankruptcy May 31, 1974; the accountants' report for the fiscal year-end
February28, 1974 is dated July19, 1974. In the previousyear,the report
was dated April 24, 1973. Note that the lead time between fiscal year
date of last "relevant" reportand date of bankruptcyis approximately
thirteenmonthsin this case (i.e., April24, 1973 to May 31, 1974). In 1974,
it took approximatelyfourand one-halfmonthsto complete the audit.
(I found many cases which exceeded fourand one-halfmonths.) There
were also a numberof firmsforwhich additional relevantreportscould
have existed. Under such circumstances,search procedures were attempted,but with littlesuccess. For most of these cases, it appears as if
the firmssimplynever filedany additional reportswith the S.E.C. This
is by no means implausible,since firmscan apply forextensionof their
deadline, and afterbankruptcyhas actually occurredthere may simply
be no point in goingthroughan audit and preparinga standard annual
report.Of course, it is also possible that additional reportsdid exist,but
never got to the StanfordUniversityLibrary.In orderto play it "safe,"
I decided that no firmwas to be deleted because reportswere potentially
missing.As a consequence, any evaluation of a model based on this data
set probablyunderstatesthe predictiveclassificationperformance.
A sample ofnonbankruptfirmswas obtainedfromthe Compustattape.
Ideally, all reports for all firmssatisfyingthe population constraints
should have been includedas a controlsample. However,thiswas deemed
to be too costly and impractical (due to core memory constraints).I
decided instead that everyfirmon the Compustat tape (excludingutilities, etc.) should contributewithonly one vectorof data points;the year
of any givenfirm'sreportwas obtained by randomprocedure.This led to
2,058 vectors of data points fornonbankruptfirms.The breakdowninto
exchange listings was as follows: New York Stock Exchange = 42%,
American Stock Exchange = 32%, Other = 26%.

4. A Probabilistic Model ofBankruptcy

Let Xi denote a vectorof predictorsforthe ith observation;let /3be a
vector of unknownparameters,and let P(X1, /8)denote the probability
of bankruptcyforany given Xi and /1.P is some probabilityfunction,0
< P < 1. The logarithmof the likelihood of any specificoutcomes, as
reflectedby the binary sample space of bankruptcyversus nonbankruptcy,is then given by:

l(B3) ieS

logP(Xf,,i) + E log(1 - P(Xi,f)),

'

ieS,

where Si is the (index) set of bankrupt firmsand

S2

is the set of

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

118

JAMES

A. OHLSON

nonbankruptfirms.For any specifiedfunctionP, the maximumlikelihood

estimatesof/1, 82 **, are obtainedbysolving:
maxl(/).
/3

In the absence of a positive theory of bankruptcy,there is no easy

solution to the problem of selectingan appropriateclass of functionsP.
As a practical matter, all one can do is to choose on the basis of
computational and interpretativesimplicity.One such functionis the
logisticfunction:
P = (1 + exp(-yi}'),

whereyi =

/,8Xij= f/'Xi.

There are two implications which should be mentioned. First, P is

increasingin y; second, y is equal to log[P/(1 - P)]. The model is thus
relativelyeasy to interpret,and thisis its main (and perhaps only)virtue.'
5. Ratios and Basic Results
For purposes of the present report,no attemptwas made to develop
any "new or exotic" ratios. The criterionfor choosing among different
predictorswas simplicity.The firstthree models estimated,Models 1-3,
were composed of an intercept and the followingnine independent
variables:10
1. SIZE = log(total assets/GNP price-levelindex). The index assumes a base value of 100 for 1968. Total assets are as reported
in dollars.The indexyear is as ofthe year priorto the year ofthe
balance sheet date. The procedureassures a real-timeimplementation of the model. The log transformhas an importantimplication. Suppose two firms,A and B, have a balance sheet date in
the same year, then the sign of PA- PB is independentof the
price-levelindex. (This will not followunless the log transformis
applied.) The latteris, of course,a desirable property.
2. TLTA = Total liabilitiesdivided by total assets.
3. WCTA Workingcapital divided by total assets.
Currentliabilitiesdivided by currentassets.
4. CLCA
5. OENEG = One if total liabilities exceeds total assets, zero
otherwise.
Net income divided by total assets.
6. NITA
7. FUTL = Funds provided by operations divided by total liabilities.
'See McFadden [1973] fora comprehensiveanalysis of the logitmodel.
10No attemptwas made to select predictorson the basis of rigoroustheory.To put it
mildly,the state of the art seems to preclude such an approach. (The firstsix predictors
werepartiallyselected simplybecause theyappear to be the ones mostfrequentlymentioned
in the literature.)

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

119

RATIOS AND BANKRUPTCY

8. INTWO = One ifnet incomewas negativeforthe last two years,

zero otherwise.
9. CHIN(NIt - NIti)/(INItI + INIt-il), where NIt is net
income for the most recent period. The denominatoracts as a
level indicator.The variable is thus intendedto measure change
in net income. (The measure appears to be due to McKibben
[1972]).
Previous studies, "common sense," and perhaps even theory,would
ratios should be
suggest that the sign of the coefficientsof the different
as follows:
Positive

TLTA
CLCA
INTWO

Indeterminate

Negative

SIZE
WCTA
NITA
FUTL
CHIN

OENEG

OENEG serves as a discontinuitycorrectionfor TLTA. A corporation

which has a negativebook value is a special case. Survival would tend to
depend upon many complicated factors,and the effectof the extreme
leverage position needs to be corrected.A positive sign would suggest
almost certainbankruptcy,while a negativesign suggeststhat the situation is verybad indeed (due to TLTA), but not that bad. (Granted,this
is a veryheuristicprocedureto capture somethingverycomplicated.)
A "profile"analysis of the data supportsthe hypothesesregardingthe
signs.Table 3 shows the means and standarddeviationsof the predictors
firms,
forthreesets of data: one year priorto bankruptcy,nonbankruptcy
and two years priorto bankruptcy.The resultsare hardlysurprising.The
ratios deteriorateas one moves fromnonbankruptfirmsto two years
priorto bankruptcyto one year priorto bankruptcy.Althoughthe data
TABLE
3
ProfileAnalysis
One Year Prior to

Bankruptcy

Variable

SIZE .............
TLTA ..............
WCTA ..............
CLCA ...............
NITA ...............
FUTL
..............
INTWO
OENEG

.............
............

CHIN ...............

N .105

NonbankruptFirs

NnakptFrsruptcy

Two Years Prior to Bank-

mean

stdv

mean

stdv

mean

stdv

12.134
0.905
0.041
1.32
-0.208
-0.117

1.38
0.637
0.608
2.52
0.411
0.421

13.26
0.488
0.310
0.525
0.0526
0.2806

1.570
0.181
0.182
0.740
0.0756
0.360

12.234
0.718
0.157
0.814
-0.052
-0.0096

1.414
0.311
0.320
0.671
0.155
0.332

0.390
0.18

0.488
0.385

0.0432
0.0044

0.2034
0.0660

0.180
0.060

0.384
0.237

-0.322

0.644

0.0379

2,058

0.458

0.00308

0.8673

100

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

120

JAMES

A. OHLSON

and ratios are not quite comparable with those of Beaver [1966], the
resultshere are quite similarto the profileshe presented[1966,p. 82]. It
should also be noted that the standard deviations of the predictors
(except for size) are larger for year-1 firms,compared to nonbankrupt
firms.These differencesare significantat a 5-percentlevel or better.
Hence, as discussed in Section 2, standard assumptions of MDA are
unlikelyto be valid.
Three sets of estimates were computed forthe logit model using the
predictorspreviouslydescribed.Model 1 predictsbankruptcywithinone
year; Model 2 predicts bankruptcywithin two years, given that the
company did not fail within the subsequent year; Model 3 predicts
bankruptcywithinone or two years.A summaryofthe resultsare shown
in table 4. This table indicates that all of the signswere as predictedfor
Model 1. Only three of the coefficients(WCTA, CLCA, and INTWO)
have t-statisticsless than two,so the othersare all statisticallysignificant
at a respectable level. This includesSIZE, whichhas a relativelylarge tis givenby the likelihood
statistic.An overall measure of goodness-of-fit
ratio index. The index is similarto a R2 in the sense that it equals one in
are zero." For
case of a perfectfit,and zero ifthe estimatedcoefficients
Model 1, the ratio is 84 percent,and this is significantat an extremely
low a-level. The statistic "Percent CorrectlyPredicted" equals 96.12
percent;it is tabulated on the basis ofa cutoffpointof .5. That is, classify
the company if and only if P(XI., /) > 0.5. Whetherthis is a "good" or
"bad" result is not easy to answer at this stage, so furtherdiscussion
regardingthe model's predictivepoweris postponeduntilthe nextsection.
At thispoint,we can note that ifall firmswere classifiedas nonbankrupt,
then 91.15 percent would be correctlyclassified (2,058/(105 + 2,058)).
Thus the marginal(unconditional,prior)probabilityof bankruptcyis an
importantquantity in the above type of statistic.Further,there is no
apparent reason why .5 is an appropriatecutoffpoint,since it presumes
implicitlythat the loss functionis symmetricacross the two types of
classificationerrors.
Table 5 shows the correlationcoefficientsof the estimationerrorsin
ofthe financialstate variables (variables 1-4 in
Model 1. The coefficients
the table) are uncorrelated with those of the performancevariables
and
(variables 5-9). Hence, both sets ofvariables contributesignificantly
independentlyof each other to the likelihood function.This strongly
supports the contentionthat both sets of variables are importantin
establishingthe predictiverelationship.
Models 2 and 3 have somewhatweakergoodness-of-fit
statistics,which
The likelihoodratio index is definedto be:
1 - log likelihoodat convergence/loglikelihoodat zero.
The index will take on the value of one in case of a perfectfit,since log likelihood at
convergencethen equals zero. If there is no fit,then obviouslythe index equals zero. See
McFadden [1973] forfurtherdetails.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

c"

~~

c~~~o ~

Cq

U0~I

Cl~
-CiclC~

CI

Cq

Cl1

UrCS
Lo
00
Cli

Cq

00

U-D

lN

cI

0II
o

0
%

oq~

Cq
m

lC

>1
-

m C0.0

0C0

u0

o
Cli
S~~~~~~~~~~~~~~~~~f
q-c-i
oio
oro

ro

~ ~

CZ

n
LoI-

Z?

121

Ho

C
lf

BANKRUPTCY

LO

CO

oo

't

oq

AND

RATIOS

l Lo lf

li

C
?

-e

>

'

Q0~~~~~~~~~~~~~

m~~~~~

00

wr
C

LL

KL

E-4~~~~~~~~~~~~~>

O. CS

c0
0 Q

cn CC
Q0

cc cc
.

>

cs

0t

.:

cc
.

C Z

CIL

cn CC

cn CC

c\)~00

cc

cc

C CJ
Z

C4

-IL

cc o

~ ~
~

C Cio
Z

-IL

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

122

JAMES A. OHLSON
TABLE
5
CorrelationMatrix of (Estimated) Estimation Errors in Model 1
SIZE

SIZE .......
TLTA ......

WCTA

CLCA
OENEG

TLTA

WCTA

-.28
1

.32

-.49

CLCA

.46

NITA
FUTL

INTWO....
CHIN
* = Absolute
valueofcoefficient
lessthan.20.

OENEG

NITA

FUTL

INTWO

CHIN

-.41
1

.40
*

-.44
*

-.32
1

is exactlywhat one would expectin view ofthe profileanalysis.Note also

that the signs of INTWO and CHIN differfromthose of Model 1. The
positive and significantcoefficientforCHIN in Model 2 can perhaps be
explained by a scenarioproposed by Deakin [1972]. Firmswitha positive
change in earningsmay be particularlytemptedto raise externalcapital
throughborrowing,and thiswillthenimplythat theybecome higher-risk
firmsat a subsequent point. Of course,this is only one possible explanation and the evidence is fartoo weak even vaguelyto suggestthat it is in
factthe case.
In all threemodels,size appears as an importantpredictor.This finding
is consistentwithHorrigan's[1968] studyofbond ratings,whereinhe too
foundthat size was an importantdeterminant.One could perhaps argue
that the conclusion is invalid because the bankruptand nonbankrupt
firmsare drawn fromdifferentpopulations. Specifically,one cannot be
sure that all the nonbankruptfirmswould have been on the Compustat
tape if they had not failed.12 No directtest of the problem is therefore
feasible.The Compustattape is heavilybiased towardthe relativelylarge
firmslisted on the two major exchanges. If size is a spurious variable,
then it is likely that dummy variables reflectingexchange listingsare'
more importantthan size. This testwas implemented,and the resultsare
shown in table 6. The t-statisticforSIZE is largerthan two,whereas the
two exchange dummies NYSE and AMSE are essentiallyinsignificant.
These results again support the contentionthat size is an important
predictorofbankruptcy.Even so, the testand conclusionmustbe viewed
fromnon-Comwith great caution, since Compustat firmsare different
pustat firmson a numberof complex dimensions.Therefore,we cannot
be sure that size is a surrogatewhich is superiorto exchange listing.If
size, in fact,is a superiorsurrogate,thenthe statisticalsignificanceofsize
may simplyreflecta general Compustatbias.13
12 The Compustat file does not include all firmswhich have (or have had) theirequity
traded.
13 It would have been preferableto obtain data fornonbankrupt
firmsby sampling10-K
reportsfromthe StanfordLibrary.However,this would have been a verycostlyprocedure.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL RATIOS AND BANKRUPTCY

123

TABLE
6
Model 4
Variable

Estimates

SIZE ..........
-.267
TLTA ...
5.63
.
WCTA .........
-1.43
CLCA .. .........
.0585
NITA ....-...
2.35
FUTL
-1.99
...........
INTWO .......
.307
-1.56
OENEG .........
CHIN .....
. -.
5092
NYSE ........
-.854
AMSW
-.0513
.......
CONST
-2.63
..........
Percent CorrectlyPredicted
Likelihood Ratio Index

t-Statistics
-2.02
6.04
-1.91
.595
-1.82
-2.53
.877
-2.20
-2.15
-1.71
-.186
-1.70
96.30%
.8399

Subject to the qualificationabove, the results indicate that the four

factorsderived fromfinancialstatementswhich are statisticallysignificant forpurposes of assessing the probabilityof bankruptcyare: (i) size
(SIZE); (ii) the financialstructureas reflectedby a measure of leverage
(TLTA); (iii) some performancemeasure or combinationofperformance
measures (NITA and/orFUTL); (iv) some measure(s) of currentliquidity (WCTA or WCTA and CLCA jointly).14 In an attemptto determine
whetherother factorscan be obtained fromfinancialstatementswhich
could increase the predictabilityof failure,I estimated an additional
model. This model was Model 1 supplementedby a measure of profit
margin,computed as fundsfromoperationsdivided by sales, and a ratio
of assets with little or no cash value (intangiblesplus deferredassets)
divided by total assets. The estimationresultsshowed not onlythat both
but also that the estiof these variables were completelyinsignificant,
mated coefficientshad "incorrect"signs. (The t-statisticswere .14 and
-.42, respectively.)Whetherotheraccountingpredictorscould have done
a betterjob in improvingupon the likelihoodfunctionis not clear,but,in
On the otherhand, nonaccountingdata,
my view,it is not overlylikely.15
such as informationbased on equity prices and changes in prices might
prove to be most useful. I intend to test these kinds of models once
adequate data have been gathered.
" If CLCA is deleted in the equation estimated,then the t-statisticof WCTA is slightly
greaterthan two. For all practicalpurposes,this will be the onlydifference.
" One possible exception would be measures of "trends" of ratios and/or volatilityof
ratios (and performancemeasures). Due to the considerablecosts of gatheringand organizing data, such a refinementwas judged to be cost-benefitinefficient.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

124

JAMES

A. OHLSON

6. Evaluation of PredictivePerformance
There is no way one can completelyorderthe predictivepowerof a set
of models used for predictive(decision) purposes. As a minimum,this
requires a complete specificationof the decision problem,includinga
preferencestructuredefinedover the appropriatestate-space. Previous
work in the area of bankruptcypredictionhas generallybeen based on
two highlyspecificand restrictiveassumptionswhen predictiveperformance is evaluated. First,a (mis)classificationmatrixis assumed to be an
adequate partition of the payoffstructure.Second, the two types of
classificationerrorshave an additive property,and the "best" model is
one which minimizes the sums of percentage errors. Both of these
assumptions are arbitrary,although it must be admitted that the first
assumptionis of some value if one is to describeat least one implication
of using a model. Much of this discussionwill thereforefocus on such a
(mis)classificationdescription.Nevertheless,the second assumptionwill
also be used at some points,since it would otherwisebe impossibleto
compare the resultshere withthose of previousstudies.The comparison
cannot be across models because the time periods,predictors,and data
sets are different.Rather, the question of interestis one of findingto
what extentthe resultsconformwith each other.
Withoutloss ofgenerality,one may regardthe estimatedprobabilityof
failure,PMX1,fi),as a signal which classifiesfirmi into one of the two
groups.Hence, it is ofinterestto describethe conditionaldistributionsof
these signals. Figure 1 shows the empiricalfrequencyofP(X,, /3)forthe
105 firms which failed within a year; fiis the vector of estimated
coefficientsobtained fromModel 1. Figure2 shows the frequencyforthe
2,058nonfailedfirmswhere/3is again takenfromModel 1. Figure3 shows
the probabilitiesfor 100 firmstwo periods priorto bankruptcy;the /P's
Occurrence (percentage)
10.0

8.0

60
....

....

40

0. .

0
5.0.0.0

... ... ... ...

20

30

FI.
.-irs
neyarprortobakupcy(15.irs

90

rooiltyofBnkupc

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL
Occurrence

RATIOS

AND

125

BANKRUPTCY

(percentage)

80

60

40

20

.10

.20

.30

.40

.50

.60

.70

80

90

I 00

Probability of Bankruptcy

FIG. 2.-Nonbankrupt firms(2,058firms)

Occurrence (percentage)
2.0

1.0

I.

..

..

:..:

::::: ::::
_
,,:
...:.......
....m..
..
. ....,
..'.

05.. . . . . . . . .

..

.. O..
... ..

10

.40

.50

60

.70

.80

.90

1 00

Probabilityof Bankruptcy

FIG. 3.-Firms two years priorto bankruptcy(100 firms)

were taken fromModel 2. The mean probabilitiesare .39, .03, and .20,
respectively.This is, ofcourse,in accordance withwhat one would expect
on the basis of priorreasoning.
Using the data which underlie figures1-3, one can readily perform
analysis of classificationerrorsfordifferent
cutoffpoints.The focuswill
be on a predictionof bankruptcywithinone year. A Type-I errorwill be
said to occur if P(X1, fi)is greaterthan the cutoffpoint and the firmi~s
nonbankrupt;in a similarfashion,one definesa Type-II errorforbankrupt firmsif the probabilityis less than the cutoffpoint. It would have
been preferableto performthe erroranalysis on a "fresh"data set and
thereby (in)validate the models estimnated.16
Due to the lack of data
beyond 1976,this was not possible at the time of the study.This should
not be a serious problem,however,forthe followingreasons.First,I have
'lbIt would, of course, have been possible to cut the sample in half and then go through
the usual kind of procedures. However, the primarypurpose of this paper is not one of
gettinga preciseevaluation ofa predictivemodel. Hence, it was decided thatthe fullsample
would be used in orderto produce the smallestpossible errorsofthe estimatedcoefficients.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

126

JAMES

A. OHLSON

Type I (percentage)
100r

FREQUENCY OF ERROR

80 -

60

40

20

0
0

20

40

60

80

100

Type IT (percentage)
FIG. 4.

not indulgedin any "data dredging"and no attemptwas made to finda

"best" model or even a model which is "superior"to Model 1. Second, a
logit analysis is not an econometricmethodwhichis designedto findan
"optimal" frontier,
tradingoffone type of erroragainst another.This is
in contrast to MDA models which satisfyoptimalityconditionsunder
appropriateassumptions.Third,as it turnsout,the sum ofthe percentage
of errorsis relativelyrobustacross a wide range of cutoffpoints.Finally,
the sample size is relativelylarge, so the estimates should not be too
sensitiveto the particularsample used."7
Figure4 depictsthe frontiertradingofone erroragainst another,when
the errorsare expressedas percentages.Figure5 showsthe mappingfrom
cutoffpoints to the two different
typesof errors.The cutoffpoint which
minimizesthe sum of errorsis .038. At that point, 17.4 percent of the
nonbankrupt firms and 12.4 percent of the 105 bankrupt firms are
misclassified.Given an (infinite)population,in which half of the firms
are failedand the otherhalfnonfailed,the expectederrorrate would then
be 14.9 percent,providedthat the cutoffpoint is .038,and providedthat
the distributionof the predictorsis representativeof the population. I
mightalso note that if one selects a cutoffpoint equal to .0095,then no
Type-II erroroccurs and Type-I errorsequal .47.
It is not easy to compare the results above with those reported by
others. First, the lead time from last fiscal year-end to the date of
17 The pointsjust made will be invalid in a real-worldapplicationof the model ifthe ,8in futureperiods.This is a fairlyobvious observation
different
parametersare significantly
which has been made by others (see, e.g., Moyer [1977]).

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

127

BANKRUPTCY

Cut Off Point (Pc)

(percentage)
_;100

_-80

Type I

Type II

60

40

20

60

40

20

20

40

60

80

100

Frequency of Error (percentage)

FiG. 5.

bankruptcyis much longer in the present study. Second, most studies,

except forthe ones by Altman and McGough [1974], Moyer [1977], and
Altman et al. [1977], have not considereddata fromthe seventies.Third,
studies have differedin theirselection of predictors.However,it is hard
to believe that the latter is of great significanceonce a fairnumberof
"basic" ratios and predictorsare incorporated.One potentiallyimportant
exception to the latter is the use of nonaccounting-baseddata such as
market-pricedata. Such data have not been used here. Fourth, it is
possible that the predictiveresults could be sensitiveto choice of estimation procedures (e.g., the logitmodel versusMDA).
A reviewofpreviousstudiesindicatesthat most ofthese reportederror
rates which are less than those givenhere,withseveral studiesreporting
rates in the vicinityof 5 percent.Some ofthe potentialsources that may
are consideredbelow.
account forthis differential
The firstpoint mentioned(the lead time) can be investigatedto some
extent. If it is true that this is an importantconsideration,then the
average lead time should be larger for misclassifiedbankrupt firms,
compared to correctlyclassified bankruptfirms.The differenceis approximately1.75 months,so the factormay be of some importance.The
differenceis significantat the 20-percentlevel (approximately).Indeed,
three of the misclassifiedfirmswould have been "correctly"classifiedif
I had used data fromthe subsequent annual report (i.e., these firms
belong to the group of eighteenfirmsdiscussed in a previous section).
Under these circumstancesthe minimumof the average erroris somewhat less than 13 percent,the cutoffpoint is somewhatlarger,and both

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

128

JAMES

A. OHLSON

types of errorsare reduced. Hence, the lead time accounts forsome of

But it is hard to believe thatthisis the entireexplanation.
the differences.
As previouslyindicated, evidence exists regardingthe second issue.
Altman and McGough [1974] applied the model which Altman [1968]
developed in 1968 to twenty-eightfirmsthat failed duringthe period
1970-73. The predictorswere computed from data one year prior to
bankruptcy;the definitionof "one year priorto bankruptcy"appears to
be completelyequivalent to that of the presentstudy.In the 1968 study
Altman reporteda misclassificationrate of approximately5 percentfor
failedfirms.The same model applied to the second groupoffirmsyielded
a misclassificationrate of 18 percent (5/28). This is a substantial and
significantincrease,since the probabilityof gettingfiveor more misclassificationsout of twenty-eight
observationsis approximatelyone in one
thousandifthe trueparameteris 5 percent.The studylacks a systematic
analysis ofthe othertypeoferror,so the evidenceis not conclusive.More
important,Altman et al. [1977] reworkedmuch of what Altman did in
1968 using data from1969-75 (94 percent of the total sample was from
this period). The model developmentincluded a numberof refinements
in the utilizationof discriminantanalysis,as well as in the computation
of financial ratios. The authors report:

". . .

bankruptcy classification

accuracy ranges fromover 96 (93% holdout) one period prior to bankruptcy"(Altmanet al. [1977,p. 50]). Needless to say, such resultsare not
in accordance withthose of the presentstudyor,forthat matter,the two
otherstudies whichused data fromthe seventies.I am unable to account
Altman et al. [1977] do not reporton
forthis difference.Unfortunately,
the average lead time,so it is impossible to evaluate the importanceof
thisfactor.Also, theydid not apply (or report)the predictiveperformance
of their recentZETA model on the 1974 sample. However, the authors
do seem to suggestsome sample dependence (see Altman et al. [1977, n.
16]). There are also differencesin the definitionof bankruptcy.
In sum, differencesin results are most difficultto reconcile. Moyer
[1977] recentlyreexaminedthe Altmanmodel using data fromthe 196575 period. (The Altman [1968] sample was from the 1946-65 period.) The

errorrate reportedby Moyer forthe Altman model was no less than 25

percent! Reestimation of the parameters of the Altman model (using
1965-75 data and using the estimatedmodel to classifyfirms)yielded an
errorrate of 10 percent.The latter result must be qualified (downward
bias) due to the small sample size (twenty-sevenbankruptand twentyseven nonbankruptfirms).The data were derivedfromMoody's Manual
leading to additional downwardbias.
Would the use of otherpredictorshave affectedthe predictivepower?
The question cannot really be answered unless one tries out a large
numberof models and thereafter,necessarily,cross-validatesthe "best"
model. The resultsfromthe model whichadded two predictorsto Model
1 are not veryencouraging,so at this point I must be quite skeptical.As

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

BANKRUPTCY

129

is more
improvementin goodness-of-fit
previouslysuggested,a significant
likely to occur by augmentingthe accounting-baseddata with marketprice data.
At this point,I want to emphasize that the reportsof the misclassified
bankrupt firmsseem to lack any "warningsignals" of impedingbankruptcy.All but two of the thirteencompanies reporteda profit.The two
losses were relativelyminor(NITA was -0.022 and -0.044, respectively),
and these two companies had strongfinancialpositions (TLTA was .23
and .37, respectively).The median TLTA ratio is .55, and the range is
.23-.70. The median NITA is 3.4 percentwitha range of -0.44 to 0.156.
(The firm which had TLTA = .70 had NITA = .156.) Other ratios
analyzed showed the same "healthy" patterns.It is not surprisingthat
these firmswere misclassified,especially if one considersthe profileof
the nonbankruptfirmsshown in table 3.
Moreover, the accountants' reportswould have been of little,if any,
use. None of the misclassifiedbankrupt firmshad a "going-concern
qualification"or disclaimerof opinion.A reviewof the opinionsrevealed
that eleven of these companies had completelyclean opinions,and the
two that did not had relativelyminoruncertaintyexceptions.Curiously,
some of the firmseven paid dividends in the year priorto bankruptcy.
Hence, if any warningsignals were present,it is not clear what these
actually were.'
There is always the possibilitythat an alternativeestimatingtechnique,
other than the logit model used, could yield a more powerfuldiscrimia priorireasoningappears to be of no use
natorydevice. Unfortunately,
in findingsuch an "optimal" estimatingtechnique. All one can do is to
try some alternatives. One approach I tried, MDA, produced results
which were somewhat "worse" than those previouslyreported,in that
the minimumaverage errorrate was 16 percent.More generally,I would
hypothesizethat many"reasonable" procedureswilllead to resultswhich
will not differtoo much. This robustnesspropertycan be illustratedas
follows.Ifwe use the estimatesfromModel 2 forthe purposeofpredicting
bankruptcy within one year, the fl-estimatesfrom Model 2 will be
evaluated in terms of their predictivepower with respect to firmsone
year priorto bankruptcyand the 2,058 Compustatfirms.Again,different
cutoffpoints yield a trade-offbetween the two types of errors.Table 7
displaysthe two typesof errorsat selected cutoffpointsforModels 1 and
2. Interestinglyenough, if a cutoffpoint of .08 is selected forModel 2,
then the average erroris 14.4 percent,and this is slightlybetterthan the
minimumattained by Model 1. Model 2 performsbetterat some other
18 Ratios other than those used in the estimatingequations were also examined. For all
of the misclassifiedfirms,I was unable to detect any ratio which was clearly out of line.
However,it is quite possible that a time-seriesanalysisofan extendedperiodwould indicate
thatsome ofthe firmshad "significant"above-averageoperatingrisks.By the use ofmarket
data, this problemwill be investigatedin the future.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

JAMES

130

A. OHLSON
TABLE

Type I-Type II Analysis forSelected CutoffPoints*

Estimates from:
CutoffPoint
0.0
0.02
0.04
0.06
0.08
0.10
0.20
0.30
0.40
0.42
0.50
0.54
0.60
0.70
0.80
0.90
1.00

Model 1
Type I**
Type II

Type I

Model 2
Type II

0%
7.6
14.3
20.0
25.7
26.7
44.8
48.6
57.1
61.0
67.6
68.6
71.4
76.2
81.9
88.6
1.00

100%
54.3
37.7
26.8
20.2
17.0
7.2
3.6
2.0
1.75
1.07
0.82
0.68
0.49
0.24
0.19
0

0%
0%
0.95
4.76
8.6
12.4
31.4
43.8
50.5
51.4
57.1
61.0
62.9
70.5
74.3
82.9
1.00

100%
28.7
16.7
11.8
9.3
7.2
3.3
1.75
1.07
0.92
0.63
0.44
0.29
0.19
0.15
0.049
0

* Data sets: nonbankruptfirms and one year priorto bankruptcy.

** Type I: predictbankruptcy;actual nonbankrupt.

points too; that is, fora fixedlevel of one type of errorforboth models,
the complementaryerroris lower for Model 2. A close examinationof
table 7 will verifythis. To be sure,forsome errorrates Model 1 is better
than Model 2. It seems reasonable to suggest that the models are
essentiallyequivalent as predictivetools.

7. Conclusions
There are two conclusionswhich should be restated.First,the predictive power of any model depends upon when the information(financial
report)is assumed to be available. Some previousstudies have not been
carefulin this regard.Second, the predictivepowersof linear transforms
of a vector of ratios seem to be robust across (large sample) estimation
procedures. Hence, more than anythingelse, significantimprovement
probablyrequires additional predictors.
REFERENCES
E. "Financial Ratios, DiscriminantAnalysis and the Prediction of Corporate
Bankruptcy."Journal of Finance (September 1968).
. Corporate Bankruptcyin America. Lexington,Mass.: Heath Lexington,1971.
. "PredictingRailroad Bankruptciesin America." Bell Journal of Economics and
Management Science (Spring 1973).
, R. HALDEMAN, AND P. NARAYANAN. "ZETA Analysis:A New Model to Identify
BankruptcyRisk of Corporations."Journal ofBanking and Finance (June 1977).
Broker
, AND B. LORRIS. "A Financial Early WarningSystem forOver-the-Counter
Dealers." Journal of Finance (September 1976).

ALTMAN,

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions

FINANCIAL

RATIOS

AND

BANKRUPTCY

131

, AND T. McGOUGH. "Evaluation of a Company as a Going Concern." Journal of

Accountancy (December 1974).
BEAVER, W. "Financial Ratios as Predictorsof Failure." Empirical Research in Accounting: Selected Studies, 1966. Supplementto Journal ofAccountingResearch 4.
"AlternativeFinancial Ratios as Predictionsof Failure." The AccountingReview
(January1968a).
. "Market Prices, Financial Ratios and the Prediction of Failure." Journal of
AccountingResearch (Autumn 1968b).
BLUM, M. "Failing Company DiscriminantAnalysis." Journal of Accounting Research
(Spring 1974).
DEAKIN, E. B. "A DiscriminantAnalysis of Predictorsof Business Failure." Journal of
AccountingResearch (Spring 1972).
EDMISTER,R. 0. "An EmpiricalTest ofFinancial Ratio AnalysisforSmnallBusiness Failure
Prediction."Journal ofFinancial and QuantitativeAnalysis (March 1972).
R. A. "Pitfallsin the ApplicationofDiscriminantAnalysisin Business,Finance,
EISENBEIS,
and Economics." Journal of Finance (June 1977).
HORRIGAN,J. "The Determinationof Long-TermCredit StandingwithFinancial Ratios."
Empirical Research in Accounting: Selected Studies, 1968. Supplement to Journal of
AccountingResearch 6.
JoY,M. O., AND J. 0. TOLLEFSON. "On the Financial ApplicationsofDiscriminantAnalysis."
Journal of Financial and QuantitativeAnalysis (December 1975).
LEV, B. "Financial Failure and InformationalDecomposition Measures." In Accountingin
Perspective Contributionsto AccountingThoughtsbyOtherDisciplines, edited by R. R.
Sterlingand W. F. Bentz, pp. 102-11. Cincinnati:SouthwesternPublishingCo., 1971.
LIBBY, R. "AccountingRatios and the Predictionof Failure: Some Behavioral Evidence."
Journal ofAccountingResearch (Spring 1975).
D. "ConditionalLogit Analysis of Qualitative Choice Behavior." In Frontiers
McFADDEN,
in Econometrics,edited by P. Zarembka. New York: Academic Press, 1973.
W. "EconometricForecastingof Common Stock InvestmentReturns:A New
McKIBBEN,
MethodologyUsing Fundamental OperatingData." Journal ofFinance (May 1972).
MOYER, R. "Forecasting Financial Failure: A Re-Examination." Financial Management
(Spring 1977).
SANTOMERO,
A., AND J. D. VINSO. "Estimatingthe Probabilityof Failure forCommercial
Banks and the Banking System." Journal of Banking and Finance (September 1977).
WHITE,R. W., AND M. TURNBULL. "The Probabilityof Bankruptcy:AmericanRailroads."
Workingpaper,InstituteofFinance and Accounting,London UniversityGraduate School
of Business, February 1975a.
. "The Probabilityof BankruptcyforAmerican IndustrialFirms." Workingpaper,
July 1975b.
WILCOX,J. "A PredictionofBusiness Failure UsingAccountingData." Empirical Research
in Accounting: Selected Studies, 1973. Supplement to Journal ofAccountingResearch
11.

This content downloaded from 59.165.151.3 on Thu, 09 Jul 2015 17:52:04 UTC
All use subject to JSTOR Terms and Conditions