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

Accounting Research Center, Booth School of Business, University of Chicago
Financial Ratios As Predictors of Failure

Author(s): William H. Beaver
Reviewed work(s):
Source: Journal of Accounting Research, Vol. 4, Empirical Research in Accounting: Selected
Studies 1966 (1966), pp. 71-111
Published by: Wiley on behalf of Accounting Research Center, Booth School of Business, University of
Chicago
Stable URL: http://www.jstor.org/stable/2490171 .
Accessed: 12/12/2012 16:20
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 on Wed, 12 Dec 2012 16:20:50 PM

All use subject to JSTOR Terms and Conditions
FinancialRatios as Predictorsof Failure
WILLIAM H. BEAVER*
At the turnof the century,ratio analysis was in its embryonicstate. It

began with the developmentof a single ratio, the currentratio,' for a
singlepurpose-the evaluation of credit-worthiness. Today ratio analysis
involvesthe use of several ratios by a variety of users-including credit
lenders,credit-ratingagencies,investors,and management.2In spite of
the ubiquity of ratios, little efforthas been directedtoward the formal
empiricalverificationof theirusefulness.
The usefulnessof ratios can only be tested with regard to some par-
ticular purpose. The purpose chosen here was the predictionof failure,
since ratios are currentlyin widespreaduse as predictorsof failure.This
is not the only possible use of ratios but is a startingpoint fromwhich
to build an empiricalverificationof ratio analysis.
"Failure" is definedas the inabilityof a firmto pay its financialobli-
gations as theymature.Operationally,a firmis said to have failed when
any of the followingevents have occurred: bankruptcy,bond default,
an overdrawnbank account, or nonpaymentof a preferredstock divi-
dend.3A "financialratio" is a quotientof two numbers,wherebothnum-
* Assistant Professorof Accounting,Universityof Chicago.
1 The author is deeply indebted to Harry Roberts, Professorof Statistics,University
of Chicago, forhis many helpful suggestionsdealing with the statistical problems that
were encountered in this study.
2 The currentratio was cited in the literature as early as 1908: William M. Rosen-
dale, "Credit Department Methods," Bankers' Magazine (1908), pp. 183-84. For an
excellent review of the historyand uses of ratios: James 0. Horrigan, "An Evaluation
of the Usage of Ratios in External Financial Statement Analysis" (unpublished dis-
sertation proposal, Graduate School of Business, The University of Chicago, 1963).
'Of the 79 failed firmsstudied, 59 were bankrupt; 16 involved nonpayment of pre-
ferredstock dividends; 3 were bond defaults; and 1 was an overdrawn bank account.
For a more complete discussion of the various definitionsof failure: Charles W. Ger-
stenberg, Financial Organization and Management of Business (3rd ed., rev.; New
York: Prentice-Hall, Inc., 1951), p. 569.
71

72 EMPIRICAL RESEARCH IN ACCOUNTING: SELECTED STUDIES, 1966
bers consistof financialstatementitems.A thirdterm,predictiveability,

also requires explanation but cannot be defined briefly.The various
dimensionsofpredictiveabilitywill be exploredlater.
The emphasisupon financialratios does not implythat ratios are the
only predictorsof failure.The primaryconcernis not with predictorsof
failureper se but ratherwith financialratios as predictorsof important
events-one of whichis failureof the firm.Further,the primaryconcern
is not withthe ratios as a formof presentingfinancial-statement data but
ratherwith the underlyingpredictiveability of the financialstatements
themselves.The ultimatemotivationis to provide an empiricalverifica-
tion of the usefulness (i.e., the predictiveability) of accountingdata
(i.e., financialstatements).
The study is intendedto be a test of the status quo. It is based upon
financialstatementspreparedunder currentlyaccepted reportingstand-
ards, and it employs the currentlyaccepted form for analyzing the
statements-financialratios. The study is offerednot as one of the last
endeavorsin this area but as one of the first.It is designedto be a bench-
mark for futureinvestigationsinto alternativepredictorsof failure,into
competingforms of presentingaccounting data, and into other uses
foraccountingdata.
The paper consistsof fiveparts: (1) the sample design,(2) a compari-
son of means, (3) a dichotomousclassificationtest, (4) an analysis of
likelihoodratios,and (5) concludingremarks.
Selection of Failed Firms

The most difficulttask of data collection was findinga sample of
failed firmsforwhich financialstatementscould be obtained. Afterthis
problem was discussed with representativesof several organizations
who mighthave access to the information, it was decided that Moody's
Industrial Manual was apparentlythe only source available.
Moody's Industrial Manual containsthe financial-statement data for
industrial,publiclyowned corporations.The populationexcludesfirmsof
noncorporateform,privatelyheld corporations,and nonindustrialfirms
(e.g., public utilities,transportationcompanies, and financial institu-
tions). The firmsin Moody's tend to be largerin termsof total assets
than are noncorporatefirmsand privately held corporations.Strictly
speaking,inferencesdrawn fromthis study apply only to firmsthat are
membersof the population.
The choice of this population is admittedly a reluctant one. The
probabilityof failure among this group of firmsis not so high as it is
among smallerfirms.In this sense,it is not the most relevantpopulation
upon which to test the predictiveability of ratios. However, the situa-
tion may not be so bleak as it firstappears. The populationchosenis not
a trivial one; it representsover 90 per cent of the investedcapital of all

FINANCIAL RATIOS AS PREDICTORS OF FAILURE 73
industrialfirms-investedcapital representingthe contributionsof sev-

eral millioninvestorsand creditors.
The findingsmay also possess some relevance for smaller firmsas
well. The crucialquestionis-does the Moody's populationdifferin some
essential way fromthat of the smaller firms?The question cannot be
directlyansweredwithinthe contextof the study. The answer can only
be providedby a replicationon otherpopulations.However,the findings
here can provide some evidence regardingthe extentto which such a
replicationwould be worthwhile.
The next task was to identifythose firmsin Moody's that had failed
duringthe time period being studied (1954 to 1964, inclusive). In the
frontof Moody's there appears a list of firms-firmson whom Moody's
has formerly reportedbut no longerdoes so. There are many reasonswhy
a firm;may be dropped-name change,merger,liquidation,lack of public
interest,and, most importantly,failure. The list of several thousand
names was condensedinto a list of firmsthat had failed. Supplementing
this basic list was a list of bankruptfirmsprovidedby Dun and Brad-
street.The finallist of failed firmscontained79 firmson whichfinancial-
statementdata could be obtainedforthe firstyear beforefailure.4
Next, the failed firmswere classifiedaccordingto industryand asset
size. Each firmwas assigned a three-digitnumber that denoted its
principalline of activity; the numberingsystemused was the Standard
Industrial Classification(SIC) systemof the United States Department
of Commerce.The total asset size of each firmwas obtained fromthe
mostrecentbalance sheet priorto the date of failure (the data were col-
lected fromMoody's). The industryand asset-size compositionwere
heterogeneous.The 79 failed firmsoperated in 38 differentindustries.
Eighteen of the industriescontainedonly one failed firm;the most fre-
quently representedindustrywas Manufacturersof Electronic Equip-
ment (SIC number,367), which consisted of six firms.The asset-size
range was .6 millionto 45 million dollars, and the mean asset size was
approximately6 million.5
Selection of Nonfailed Firms

The classificationof the failed firmsaccordingto industryand asset
size was an essential prerequisiteto the selectionof the nonfailed firms.
The selectionprocess was based upon a paired-sampledesign-that is,
foreach failed firmin the sample, a nonfailedfirmof the same industry
and asset size was selected. The names of the nonfailedfirmswere ob-
tained from a list of 12,000 firms.6The list possessed two convenient
4A frequencydistribution
of the fiscalyearof failureappearsin Table A-1 in the
Appendix.
'The industrialcompositionof the sample appearsin Table A-2 in the Appendix.
12,000 Leading U.S. Corporations (New York: News Front Magazine). The list

74 WILLIAM H. BEAVER
features: (1) the firmswere grouped accordingto industry (using the

SIC system) and (2) withineach industrygroup,the firmswere ranked
accordingto size.
The followingprocedurewas used to select the nonfailedfirms: (1)
take the industrynumberof a failed firm; (2) turn to that portionof
the list where firmsof that industrynumber are found; (3) within
that industrygroup,tentativelyselect the firmwhose asset size is closest
to the asset size of the failed firm; (4) if that firmis in Moody's and is
nonfailed,then accept it as one of the firmsto be studied; (5) if the firm
does not fulfillthese two conditions,returnto the list and tentatively
select the firmwhich is next closest in asset size; (6) repeat this proce-
dure until an acceptable candidate is found; and (7) repeat this pro-
cedureforeach of the failedfirmsin the sample.
An adequate explanationof the motivationbehind the paired-sample
designinvolvesa briefdiscussionof the historyofthisstudy.The findings
presentedhere are an extensionof an earlier effortbased on the same
body of data and the same sampledesign.7
The paired-sampledesign was selected in the earlier study to help
provide a "control"over factorsthat otherwisemightblur the relation-
ship between ratios and failure.As early as 1923, the ratio literature
suggestedthat industryfactors must be incorporatedin any complete
ratio analysis.8 The literaturecontendsthat "differences"exist among
industriesthat prevent the direct comparison of firmsfrom different
industries.Anotherway of statingthis argumentis to say that the same
numericalvalue of a ratio (e.g., a currentratio of 2.00) impliesa differ-
ent probabilityof failure in differentindustries.The evidence offered
in behalf of industrydifferences is the fact that the ratio distributions
differamong industries.9However, such evidenceis not conclusive,since
the failure rate may vary among industriesto compensatefor the dif-
ferencesin the ratios. No evidence has been offeredto indicate whether
or notthe compensatingdifferences exist.
Althoughless attentionhas been directedtowardthe influenceof asset
size, there are certain statistical reasons for believing that asset size
alters the relationshipbetweenratios and failure.It can be argued that
the largerof two firmswill have a lower probabilityof failure,even if
foryears prior to 1959 was not available. The nonfailed mates of the firmsthat failed
prior to 1959 tend to be older than their failed mates. There may be a possible bias
here, but the evidence indicates that the ability to predict failure was unrelated to
fiscal year before failure, and the ratio behavior of the nonfailed firmsis quite stable
and apparently independent of age.
'William H. Beaver, "Financial Ratios as Predictors of Failure" (unpublished Ph.D.
dissertation,Graduate School of Business, Universityof Chicago, 1965).
8James H. Bliss, Financial and Operating Ratios in Management (New York: The
Ronald Press, 1923), p. 59.
'A. C. Littleton,"The 2 to 1 Ratio Analyzed," CertifiedPublic Accountant, August,
1926, pp. 244-46.

the ratios of the two firmsare identical (i.e., even if the ratios have the
same numericalvalues).
If firmsare viewed as aggregatesof assets and if asset returnsare less
than perfectlycorrelatedwith one another,statistical formulaesuggest
that the variability of total returnto the firmwill increase less than
proportionately to the size of the firm.10 The rate of returnto the firm
will become more stable as asset size increases.Empirical evidence indi-
cates that the variabilityof rate of returndoes behave in this manner.1"
The implicationis that larger firmsare more solvent,even if the value
of theirratios is the same as that of smallerfirms.12 Simply stated, the
ratios of firmsfromdifferent asset-size classes cannot be directlycom-
pared.
The paired-sampledesign in conjunctionwith the paired analysis is
one way of compensatingfor the effectsof industryand asset size.
Whenever ratios are directly compared, the comparison involves two
firmsfromthe same industryand asset-size class. As previouslyindi-
cated, each failed firmhas a nonfailed "mate" in the sample. In the
paired analysis, the ratios of the nonfailed firmsare subtractedfrom
the ratios of the failed firms;the analysis is then based upon the dif-
ferencesin the ratiosratherthan upon the ratios themselves.Since every
difference in the ratios always involves a comparisonof two firms(one
failed and one nonfailed) fromthe same industryand asset-size class,
the potentiallydisruptiveeffectsof industryand asset size are mitigated.
A paired design and a paired analysis were employed in the previous
study.
Althoughthe paired analysis is a legitimateapproach to investigating
predictiveability,it has one seriousdrawback.It cannot draw inferences
regardinga single observation-only about pairs of observations.The
earlier study is restrictedto such statementsas: Based upon an analy-
sis of the ratios, firmA is more (or less) solventthan firmB. However,
this is an ambiguousstatementbecause it ignoresthe obvious question:
How solventis firmB? To be less solventthan firmB may not be so bad,
if firmB is an extremelysolvent firm.On the other hand, to be more
solventthan firmB may be very bad, if firmB is extremelyinsolvent.
In a decision-makingsituation,the user of ratios wishes to make infer-
ences regardinga single firm-for example, what is the probabilityof
failureforthis firm?Such inferencesare beyondthe scope of the paired
analysis.
In this studythe paired analysis was droppedin favorof an unpaired
10:W. J. Baumol, "The Transactions Demand for Cash: An Inventory Theoretic

Approach," Quarterly Journal of Economics, November, 1962, p. 556.
" Sidney Alexander, "The Effectof Size of Manufacturing Corporation on the Dis-
tribution of the Rate of Return," Review of Economics and Statistics, August, 1949,
pp. 229-35.
19"Solvency" is definedhere in terms of probability of failure.

analysis, which does permit inferencesregardinga single observation.

The eliminationof the paired analysis does not necessarilyimply that
the paired designshould also be discarded.The problemof selectingnon-
failed firmswas re-examinedto see if the paired analysis still served
someusefulpurpose.
If the comparisonof failed and nonfailedfirmsis to be meaningful,
the sample of nonfailedfirmsshould be drawn fromthe same population
as that of the failed firms.For example,failureis virtuallynonexistent
among very large firms.A comparisonof large firmswith the failed
firmswould not be particularlymeaningful,since the two groups of
firmscome fromdifferent populations.The failed firmswere drawnfrom
a population whose asset-size range does not include the very large
firms.
A random selection of nonfailed firmswithoutregard for asset size
would be inappropriate.Anotheralternativewould be a randomselection
of firmswhose asset size is less than some specifiedvalue. However, it
would be difficultto determinewhat that value should be. The paired-
sample design offersa sampling technique for selectingnonfailedfirms
froma relevantpopulation,because it draws nonfailedfirmsonly from
those asset-size (and industry) classes where failure has actually oc-
curred. The paired-sampledesign is not the only acceptable approach
but is a convenientone,in view of the earlierstudy.
The use of the paired design limitsthe scope of inferencesin at least
one respect.It is possible that industryor asset size may be an excellent
predictorof failure. While the paired design mitigates the disruptive
influenceof the industryand asset-size factors,it also virtuallyelimi-
nates any predictivepower these factorsmay have had. Therefore,the
findingswill not provide any insightinto the potentialpredictivepower
of industryand asset size.
The use of the unpaired analysis assumes that the residual effects
of industryand asset size are not very great.If this assumptionis incor-
rect,the findingscould seriouslymisstatethe predictiveability of ratios.
Evidence will be offeredlater as to the validity of the assumption.If
some residual effectsare present,the size effectcould be incorporated
into some formof multivariateanalysis,since size is easily quantifiable.
The same is nottrueof industry,wherean analysis of its effectswould be
more complex.The problemof the effectsof industryand asset size is
complicatedin still anotherway, because the pairing for asset size was
not perfect.
The mean asset size of the failed firmsis $6.3 million,while the mean
asset size of the nonfailedwas $8.5 million.13This situationprovidesan
interestingillustrationof the subtleways in whichsystematicdifferences
can be introduced.In orderto be accepted,a nonfailedfirmmust fulfill
1 The median differences
in asset size were $4.4 million (failed), and $5.8 million
(nonfailed).

two conditions:(1) it mustbe closestin asset size to the failed firm-spe-

cial effortwas made to select the closest firm,whetherthat firmwas
smalleror largerthan the failed firm,(2) the firmmust be in Moody's.
Since the probabilitythat a firmwill be in Moody's increases as asset
size increases,an acceptable nonfailedfirmwas more likelyto be larger,
ratherthan smaller,than the failedfirm.
The implicationsof the imperfectpairingare difficult to assess. A test
will be conductedto discoverthe extentto which the imperfectpairing
biases the findingsregardingthepredictiveabilityofthe ratios.
Collection of Financial Statement Data

Financial-statementdata of the failed firms were obtained from
Moody's for five years prior to failure.The "firstyear before failure"
is definedas that year included in the most recent financialstatement
prior to the date that the firmfailed. Anotherconditionalso had to be
fulfilled-the financial statementscould not be more than six months
old at the date of failure.The "second year beforefailure"is the fiscal
year precedingthe firstyear. The third,fourth,and fifthyears are simi-
larly defined.The financialstatementsof the nonfailedfirmswere ob-
tained forthe same fiscalyears as thoseof theirfailedmates.
The financial-statement data were then grouped according to year
before failure. For example, if two firmsfailed in 1964 and 1954, re-
spectively,and theirmost recent financialstatementswere prepared on
December 31, 1963 and 1953, respectively,the firstyear beforefailure
would include the 1963 statementsof the formerand the 1953 state-
ments of the latter. The fifthyear before failure would include the
1959 and 1949 statements.Since the firmsfailed from 1954 to 1964,
inclusive,the firstyear beforefailureincludes financialstatementsfrom
an eleven-yearperiod, 1953 through1963, while the fifthyear includes
1949through1959.
The financial-statement data of the nonfailedfirmswere also strati-
fied into years beforefailure,correspondingto the years that were as-
signedto theirfailed mates. For example,the 1963 (1953) statementsof
the nonfailedmate to the first(second) firmmentionedabove would be
assignedto the firstyear beforefailure.Similarly,the 1959 (1949) state-
mentswould be assignedto the fifthyear beforefailure.
Because financial-statement data were not available for every year
beforefailure the numberof observationsdecreases as the time period
beforefailureincreases.The sample size is largestin the firstyear before
failure-158 firms-and is smallestin the fifthyear-117 firms.'4Availa-
bilityof data over the five-yearperiod was not made a criterionforin-
clusion in the sample, since a possible bias mightbe introducedin this
14 The numberof observations
in each year beforefailureappearsin Table A-3 in
theAppendix.

TABLE 1
List of Ratios Testeda
GROUP I (CASH-FLOW RATIOS) GROUP V (LIQUID-ASSET TO CUR-

1. Cash flowto sales RENT DEBT RATIOS)
2. Cash flowto total assets 1. Cash to currentliabilities
3. Cash flowto net worth 2. Quick assets to currentliabilities
4. Cash flowto total debt 3. Current ratio (current assets to
GROUP II (NET-INCOME RATIOS) currentliabilities)
1. Net incometo sales GROUP VI (TURNOVER RATIOS)
2. Net incometo total assets 1. Cash to sales
3. Net incometo net worth 2. Accountsreceivableto sales
4. Net incometo total debt 3. Inventoryto sales
GROUP III (DEBT TO TOTAL-ASSET 4. Quick assets to sales
RATIOS) 5. Currentassets to sales
1. Currentliabilitiesto total assets 6. Workingcapital to sales
2. Long-termliabilitiesto total assets 7. Net worthto sales
3. Currentplus long-term liabilitiesto 8. Total assets to sales
total assets 9. Cash interval (cash to fund ex-
4. Currentplus long-term pluspreferred pendituresforoperations)
stock to total assets 10. Defensive interval (defensiveas-
GROUP IV (LIQUID-ASSET TO sets to fund expenditures for
TOTAL-ASSET RATIOS) operations)
1. Cash to total assets 11. No-creditinterval (defensiveas-
2. Quick assets to total assets sets minus currentliabilities to
3. Currentassets to total assets fund expendituresfor operations)
4. Workingcapital to total assets
a The components ofthe ratiosare definedin thefollowingmanner:cash flow-net
incomeplus depreciation,depletion,and amortization;net worth-commonstock-
holders'equity plus deferredincometaxes; cash-cash plus marketablesecurities;
quick assets-cash plus accountsreceivable;workingcapital-currentassets minus
currentliabilities; fund expendituresfor operations-operatingexpenses minus
depreciation,depletion,and amortization;and defensiveassets-quick assets.
manner.The bias would arise in that the failed firmswould have dem-
onstratedan ability to survive for at least five years; the result might
be the eliminationof the most serious failures.
Computationof Ratios
For every set of financialstatementsavailable, 30 ratios were com-
puted.15Three criteriawere used to selectthe 30 ratios fromthe set of all
possible combinationsand permutationsof financial statementitems.
The firstcriterionwas popularity-frequentappearance in the litera-
ture.16The ratios advocated in the literatureare perceivedby many to
to see the extentto
reflectthe crucial relationships.It will be interesting
which popularitywill be self-defeating-thatis, the most popular ratios
" The 30 ratiosappearin Table 1.
'" The selectionof ratioswas based upon theirfrequencyof appearancein nine-
teenfinancial-statementanalysistexts.

will becomethose mostmanipulatedby management(an activityknown

as windowdressing)in a mannerthat destroystheirutility.
The second criterionwas that the ratios performedwell in one of the
previous studies. This criterionwill enable the study to examine the
consistencyof its findingswith those of the previousstudies. The third
criterionwas that the ratio be definedin termsof a "cash-flow"concept.
The cash-flowratios offermuch promiseforprovidingratio analysis with
a unifiedframework, yet as of now they are untested.The presenceof
any one of the criteriawas a sufficientconditionforinclusionin the study.
In every instance,the ratio selected had been previouslysuggestedin
the literature;the study restricteditselfto testingexistingratios rather
than to developingnew ones.
AlthoughI excluded any ratio that was simply a transformationof
anotherratio already selected,the ratios still had commonnumerators
or commondenominators.In a multiratioanalysis it is desirableto have
each ratio convey as much additional informationas possible-that is,
the common elements should be reduced to a minimum.Accordingly,
the 30 ratios were divided into six "commonelement"groups.Only one
ratio fromeach groupwill be selectedas a focusforthe analysis.17
The analysis of the data will be divided into three sections: (1) a
comparisonof mean values, (2) a dichotomousclassificationtest, and
(3) an analysis of likelihoodratios.
Comparisonof Mean Values

The mean of the ratios was computed for the failed firmsand for
the nonfailedfirmsin each of the years beforefailure.The comparison
of mean values shall be called a profileanalysis. A profileis definedas
an outlineor a concisesketch.In psychologythe termpersonality profile
is used to summarizehow an individualperformedon various testswhich
attemptto measure several personalitydimensions.The term has been
adopted here because it provides an accurate descriptionof the kind of
analysisthat is providedby a comparisonof means.
Profileanalysis is not a predictivetest; it is merelya convenientway
of outliningthe general relationshipsbetween the failed and nonfailed
firms.The profileanalysis also facilitatesa comparisonwith previous
studies,sincetheiranalysiswas based entirelyupon the mean values.
Theory of Ratio Analysis

The "theory"of ratio analysis is an extremelysimpleone and can best
be explained within the frameworkof a "cash-flow" (i.e., liquid-asset
17 The six ratios were selected on the basis of the lowest percentage error for their
group over the five-yearperiod. Where two ratios predicted about equally well, other
factors entered, such as frequency of appearance in the literature and superior per-
formancein previous studies. The percentage error refersto the dichotomous classifi-
cation test which is explained on pp. 83-85.

flow) model8 The purposeof introducingthe cash-flowmodel is not to

have the model develop an optimal set of ratios, but ratherto use the
modelas a vehicleforexplainingthe ratiosbeingtested.
The firmis viewed as a reservoirof liquid assets, which is supplied
by inflowsand drained by outflows.The reservoirserves as a cushionor
bufferagainst variations in the flows.The solvency of the firmcan be
definedin termsof the probabilitythat the reservoirwill be exhausted,
at which point the firmwill be unable to pay its obligations as they
mature (i.e., failure).
Four conceptsare importantin drawingthe relationshipbetweenthe
liquid-asset-flow model and the ratios.The firstis the size of the reservoir
itself. The second is the net liquid-asset flow from operations,which
measuresthe net amount of liquid assets supplied to (or drained from)
the reservoirby currentoperations.The third is the debt held by the
firmand is one measure of the potential drain upon the reservoir.The
fourthis the fund expendituresfor operations and is the amount of
liquid assets drainedfromthe reservoirby operatingexpenditures.Given
these concepts,four ceterisparibus propositionscan be stated:
(1) The largerthe reservoir,the smallerthe probabilityof failure.
(2) The larger the net liquid-asset flow from operations (i.e., cash
flow),the smallerthe probabilityof failure.
(3) The largerthe amountof debt held, the greaterthe probabilityof
failure.
(4) The larger the fund expendituresfor operations,the greaterthe
probabilityof failure.
The four propositionscan be used to formpredictionsregardingthe
mean values of six financialratios. The six ratios are cash flowto total
debt,net incometo total assets,total debt to total assets,workingcapital
to total assets, currentratios, and the no-creditinterval.' The predic-
tionsappear in Table 2.
Analysis of Evidence
The differencein the mean values is in the predicteddirectionforeach
ratio in all fiveyears beforefailure.20Failed firmsnot only have lower
cash flow than nonfailed firmsbut also have a smaller reservoirof
liquid assets. Although the failed firms have less capacity to meet
obligations,theytendto incurmoredebt than do the nonfailedfirms.
The trendline of the nonfailedfirmshas a zero slope, and the devia-
'1 A moredetaileddiscussionappearsin JamesE. Walter's,"The Determination of
TechnicalSolvency,"Journalof Business,January,1957,pp. 30-43.
1 The intervalmeasuresappear in GeorgeH. Sorter'sand GeorgeBenston's"Ap-
praisingthe DefensivePositionof the Firm: The IntervalMeasure,"The Accounting

Review,October,1960,pp. 633-40.The basis forthe selectionof the six ratioswas
givenin n. 17 on p. 14.
' See Fig. 1.

FINANCIAL RATIOS AS PREDICTORS OF FAIIAURE 81
TABLE 2
Predictionof theMean Values ofFailed and NonfailedFirms
Ratio Predictiona
Cash flowto total debtb Nonfailed > failed

Net incometo total assets Nonfailed > failed
Total debt to total assetsb Failed > nonfailed
Workingcapital to total assets Nonfailed > failed
Currentratio Nonfailed > failed
No-creditinterval Nonfailed > failed
a Nonfailed> failedis a predictionthatthemeanvalue ofthe nonfailed firmswill
be greaterthan that of the failedfirms.
b Debt is defined
as currentplus long-term stock.
liabilitiesplus preferred
tions fromthe trendline are small. Yet the deteriorationin the means of
the failed firmsis very pronouncedover the five-yearperiod.The differ-
ence in means is evident for at least five years before failure, with
the differenceincreasingas the year of failureapproaches.
The data demonstratea substantialdegreeof consistency.The evidence
overwhelmingly in the ratios of failed
suggeststhat there is a difference
and nonfailedfirms,
Comparisonof Mean AssetSize

Both the failed and nonfailedfirmsgrowin the second throughfifth
years, althoughthe growthrate of the nonfailedfirmsis greater.In the
firstyear beforefailurethe nonfailedfirmscontinueto grow while the
in asset size indi-
total assets of the failed firmsdecline. The difference
cates that the pairing for size was not perfect.The implicationof this
findingwill be examinedlater.
ComparisonwithPreviousStudies
In 1932, Fitz Patrick published a study of nineteenpairs of failed
and nonfailedfirms.21His evidence indicatedthat therewere persistent
differencesin the ratios for at least three years prior to failure. The
Winakor and Smith study of 1935 investigatedthe mean ratios of failed
firmsforten years priorto failureand found a marked deteriorationin
the mean values with the rate of deteriorationincreasing as failure
approached.22In 1942, Merwin comparedthe mean ratios of continuing
'Paul J. Fitz Patrick,"A Comparisonof Ratios of SuccessfulIndustrialEnter-

priseswithThose ofFailed Firms,"Certified Public Accountant,October,November,
and December,1932,pp. 598-605,656-62,and 727-31,respectively.
22 ArthurWinakorand RaymondF. Smith,Changesin FinancialStructureof Un-
successfulIndustrial Companies (Bureau of Business Research, Bulletin No. 51
[Urbana:University ofIllinoisPress,1935]).

*CASH FLOW NET INCOME TOTAL DEBT

TOTAL DEBT TOTAL ASSETS TOTAL ASSETS
+45- +1 .79'
+.35 .78
+.25 3 0 .652
+.15. O .58
+.05 -A.1 .51 b
-.05 / .44.
/d
-.15. /-.2 .37
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
YR. BEFORE FAILURE YR. BEFORE FAILURE YR. BEFORE FAILURE.
WORKING CAPITAL CURRENT NO CREDIT
TOTAL ASSETS RATIO INTERVAL
.42- 3.5 . +.15
.366
.30 -o c 3.0- +.05
.24 /
.11 / 2.5- -.05-
.12 /
.06 2.01(15
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
YR. BEFORE FAILURE YR. BEFORE FAILURE YR. BEFORE FAILURE
TOTAL ASSETS
(IN MILLIONS OF DOLLARS)
8F
7-
/\Q
6-
1 2 3 4 5
YR. BEFORE FAILURE
LEGEND: .--NONFAILED FIRMS - - - FAILED FIRMS
of mean values.
FIG. 1. Profileanalysis,comparisoin

firmswith those of discontinuedfirmsfor the period 1926 to 1936.23 A

differencein means was observed for as much as six years beforedis-
increasedas the year of discontinuance
continuance,while the difference
approached.The findingsof all threestudies supportthose of the profile
analysis.
Limitationsof ProfileAnalysis
Profileanalysis can demonstratethat a difference betweenfailed and
nonfailedfirmsexists,but it cannot answer the crucial question: How
large is the difference?The profileconcentratesupon a single point on
the ratio'distribution-the mean. Without the additional knowledgeof
the dispersionabout that point, no meaningfulstatementcan be made
regardingthe predictiveability of a ratio.
Consider two ratio distributions,each of which is symmetrical,and
assumethat a difference in theirmeans is observed.If the dispersionabout
themeans is tight,therecould be littleor no overlap of one distributionon
the other.Little or no overlap impliesthat the ratio would be an excel-
lent predictorof failure.However, if the dispersionabout the means is
great, there could be quite a bit of overlap, which would indicate a
lesserdegreeof predictiveabilitypresent.
If the distributionsare skewed (i.e., nonsymmetrical), the extremeob-
servations,'which constitutea small proportionof the total numberof
firms,may be responsibleformost of the difference in the means. Apart
fromtwo or three extremefirms,there could be a completeoverlap of
the distributionsof failed and nonfailedfirms.Previous studies indicate
that ratio distributions
do have pronouncedskews.24
The'precedingremarksimplythat ratios may have little or no ability
to predictfailure,in spite of the differencesin the means. The discussion
suggeststhat some sort of predictivetest is needed.25
Dichotomous ClassificationTest
The dichotomousclassificationtest predicts the failure status of a
firm,based solely upon a knowledgeof the financialratios. In contrast
to the profileanalysis,it is a predictivetest, althoughit will not provide
as much insight as likelihood ratios. Conducting this intermediate
analysis has certainadvantages.
The classificationtest is an intuitivelyappealing approach. It closely
resembles the decision-makingsituation facing many users of ratios.
3Charles L. Merwin, Financing Small Corporations in Five Manufacturing Indus-
tries,1926-36 (New York: National Bureau of Economic Research, 1942).
24 The Earning Power Ratios of Public Utility Companies (Bureau of Business Re-
search, Bulletin No. 15 [Urbana, Ill.: University of Illinois Press, 1927]).

25 This discussion also implies that the use of standard ratios (i.e., average ratios
for an industry) may have serious limitations.

For example,a bank's lendingdecision can be viewed as a dichotomous

choice of accepting or rejectinga loan application. The object of the
ratio analysis would be to classifyfirmsas acceptable or not. Of course,
the decision is not quite that simple. The bank must also decide how
much to loan and at what rate to loan, if the firmis classifiedas accept-
able. However, the example does illustratethe parallel betweencertain
kinds of decisions and the classificationtest. The test also provides a
convenientsiftingdevice forselectingthe six ratios that will serve as a
focusforthe analysis.
The classificationtest makes a dichotomousprediction-that is, a
firmis eitherfailed or nonfailed.In orderto make the predictions,the
data are arrayed (i.e., each ratio is arranged in ascending order). The
array of a given ratio is visually inspected to find an optimal cutoff
point-a point that will minimizethe per cent of incorrectpredictions.
If a firm'sratio is below (or above, as in the case of the total debt to
total-assets ratio) the cutoffpoint, the firmis classified as failed. If
the firm'sratio is above (or below,forthe total debt to total-assetsratio)
the criticalvalue, the firmis classifiedas nonfailed.
Aftereach firmhas been classified,the predictionsare comparedwith
the actual failurestatus, and the percentageof incorrectpredictionsis
computed.This procedurewas repeated for each of the 30 ratios; the
resultsforsix ratios appear in Table 3.26 The percentageof misclassifica-
tions may be taken as a crude index of predictiveability-the smaller
the error,the higherthe degreeofpredictiveabilitypresent.
The process of findingthe optimal cutoffpoint is largely one of trial
and error.The test is open to criticismon the groundsthat it selects
ex post (i.e., after looking at the data) that point which will minimize
the incorrectpredictions.In a decision-makingsituation,the user of
ratios does not have the benefitof such informationand must predicton
a new set of observations-that is, on a set of observationswhere the
actual failurestatus is not knownuntil sometimeafterthe decision has
been made.
The test was modifiedso that it morecloselyconformedto the decision-
makingsituation.The sample was randomlydividedintotwo subsamples.
An optimalcutoffpointwas foundforeach subsample,and two testswere
conducted.In the firsttest the firmsin each subsample were classified
using the cutoffpoint derived fromtheir own subsample-which is the
test outlined in the previous paragraphs. The percentageof incorrect
predictionsfor this test are the numbers in parenthesesin Table 3.
In the secondtest the firmswere classifiedusing the cutoffpoint derived
26 The
six ratios were selected on the basis of the lowest percentage error for their
group over the five-yearperiod. Where two ratios predicted about equally well, other
factors entered, such as frequency of appearance in the literature and superior per-
formance in previous studies. The results for all 30 ratios appear in Table A-4 in the
Appendix. The cutoffpoints used appear in Table A-5 in the Appendix.

TABLE 3
Percentage Dichotomous
of FirmsMisclassiJieda: Classification
Test
Year beforeFailure
Ratio
1 2 3 4 5
Cash flow .13 .21 .23 .24 .22

Total debt (.10) (.18) (.21) (.24) (.22)
Net income .13 .20 .23 .29 .28

Total assets (.12) (.15) (.22) (.28) (.25)
Total debt .19 .25 .34 .27 .28

Total assets (.19) (.24) (.28) (.24) (.27)
Workingcapital .24 .34 .33 .45 .41

Total assets (.20) (.30) (.33) (.35) (.35)
.20 .32 .36 .38 .45

Currentratio (.20) (.27) (.31) (.32) (.31)
No-creit
inerval.23 .38 .43 .38 .37
No-creditinterval (.23) (.31) (.30) (.35) (.30)
Total Assets .38 .42 .45 .49 .47

(.38) (.42) (.42) (.41) (.38)
the resultsof the secondtest. The bottomrowrefersto

a The top rowrepresents
the firsttest.
fromthe othersubsample.The second test is similarto the classification

problem facing the users of ratios. The results of the second test are
thosenumberswhichare not enclosedin parenthesesin Table 3.
Analysis of Evidence (Percentage Error)

The ability to predictfailureis strongestin the cash-flowto total-debt
ratio. Unless otherwiseindicated,the commentswill referto the second
test,whichis a moremeaningfultest. In the firstyear beforefailurethe
erroris only 13 per cent,while in the fifththe errorpercentageis 22.
Both errorpercentagesare much smallerthan would be expectedfroma
random-prediction model,wherethe expectedpercentageerrorwould be
approximately50 per cent. There is an extremelysmall probabilitythat
randompredictioncould have done as well.27
Clearly all ratios do not predictequally well. The net-incometo total-
assets ratio predictssecondbest,whichis to be expected,sinceits correla-
tion withthe best ratio is higherthan the correlationof any otherratio
27In both the firstand fifth
years,the probabilityis less than one in ten thousand.

withthe bestratio.The total-debtto total-assetsratio predictednextbest,

withthe threeliquid-assetratios performing least well. The most crucial
factorseemsto be the net liquid-assetflowsuppliedto the reservoir,while
the size of the reservoiris the least importantfactor.
The explanationof relative predictiveability is an interestingtopic,
but is not discussed here because of the additional amount of analysis
that would be required.Part of the answercan be foundin an investiga-
tion intothe behaviorof the componentsof the ratios.28
IndustryEffects
One crude test of the residual effectsof industryis the amountof in-
crease in percentageerrorfromthe firsttest to the second test,since the
industrycompositiondiffersbetweenthe two subsamples. The increase
in erroris small, especially forthe betterratios,and this findingis con-
sistentwiththe assumptionthatthe residualeffectsare not great.
Another indirecttest is the increase in percentage error from the
paired analysis to the unpaired analysis.29In most ratios the increase is
small but persistent.Althoughthe reductionin predictivepoweris by no
means overwhelming, there seems to be a small residual effectof indus-
try.To the extentthat the effectsare still present,the predictiveability
of ratios is understated.Predictiveability would be expectedto improve
if the effectof industrywere fullytaken into account.
Both tests are not directmeasures of the residual effects,because the
increasein percentageerrorscould be due to otherfactorssuch as residual
asset-size effects.The tests offerevidence of an indirectsort-that is,
nothinghas been observedto suggestthat the residualeffectsare great.
Implications of Imperfect Pairing for Asset Size

The profileanalysis indicatedthat the mean asset size of the nonfailed
firmswas greaterthan that of the failed firms.If the ratios are corre-
lated with total assets and if total assets has some explanatorypower
of its own,the findingswould overstatethe predictiveabilityof ratios.
The correlationcoefficients betweenasset size and the ratios are shown
in Table 5. The square of the coefficient(r2) can be interpretedas the
proportionof the variance of the ratio that is explainedby the variation
in total assets. In threeof the ratios,the r2 is below 1 per cent for both
subsamples,and the highestvalue of r2 is 4.8 per cent. The r2 can be
properlyinterpretedas the proportionof variance explained,only if the
relationshipbetween the variables is bivariate normal. Bivariate nor-
mality is difficultto determine,since the joint distributionmay not be
bivariate normal even if the marginal distributionsare. However, the
' Beaver, op cit., p. 46.
29 See Table 4.

TABLE 4
Comparisonof PercentageErrorsfor Paired and UnpairedClassificationTests
Year beforeFailure
Ratio
1 2 3 4 5
Cash flowto total assetsa

Unpaired .10 .20 .24 .28 .28
Paired (.10) (.13) (.20) (.23) (.28)
Difference .00 .07 .04 .05 .00

Net incometo total assets
Unpaired .13 .21 .23 .29 .28
Paired (.11) (.15) (.16) (.38) (.30)
Difference .02 .06 .07 (.09) (.02)

Total debt to total assets
Unpaired .19 .25 .34 .27 .28
Paired (.15) (.27) (.30) (.27) (.39)
Difference .04 (.02) .04 .00 (.11)

Workingcapital to total assets
Unpaired .24 .34 .33 .45 .41
Paired (.16) (.22) (.34) (.26) (.33)
Difference .08 .12 (.01) .19 .08

Currentratio
Unpaired .20 .32 .36 .38 .45
Paired (.15) (.25) (.32) (.32) (.39)
Difference .05 .07 .04 .06 .06

No-creditinterval
Unpaired .23 .38 .43 .38 .37
Paired (.21) (.38) (.43) (.43) (.43)
Difference .02 .00 .00 (.05) (.06)

a The cash-flowto total-debtratiowas not computedforthe paired analysis.The
cash-flow as a substitute.
to total-assetratio is offered
evidencepresentedin the nextsectionwill indicatethat even the marginal

distributionsare not normal,which suggeststhat inferencesbased upon
the r2 mustbe treatedwithreservation.The findingscan be interpretedin
a negativesense-at least no evidenceof strongcorrelationhas been ob-
served. The results are consistentwith the hypothesisthat the ratios
are uncorrelatedwithasset size.
The implicationof littleor no correlationis that the predictiveability
of the ratios is not overstatedby the imperfectasset-size pairing.How-
ever, if strong correlationhad been found,it would be impossible to
isolate the marginalpredictiveability of individualvariables. The mini-
mumpossible overstatement would be zero; the maximumpossiblewould

TABLE 5
Correlationof Total Assets withSix Ratios, First Year Before Failure
Correlation" Proportionof Variance

Coefficient
(r) Explained(r')
Ratio Subsample Subsample
A B A B
Cash flow .12 .20 .0144 .0400

Total debt
Net income .22 .18 .0484 .0324

Total assets
Total debt -.09 -.06 .0081 .0036

Total assets
Working capital -.15 -.01 .0225 .0001

Total assets
Current ratio -.04 .02 .0016 .0004

No-credit interval -.02 .15 .0004 .0225
a Coefficientsrounded to nearest .01. Subsamples A and B refer to the two sub-
samples discussed on p. 84.
be the predictiveabilityof total assets. Table 3 suggeststhat the residual

predictivepower of total assets is small and is virtuallynonexistentin
the thirdthroughthe fifthyears.
Ultimately,I am unconcernedabout the marginal predictiveability
of ratios versusasset size, since my primaryconcernis withthe utilityof
accountingdata regardlessof the formin which it is introducedinto
the analysis. To the extentthat total assets has predictivepower of its
own, the studyunderstatesthe predictivepower of accountingdata, be-
cause the effectof the paired designwas to reduce any predictivepower
that asset size may have had.
ContingencyTables
The percentageof incorrectpredictionssuffersfromtwo limitations.
(1) In a world wherethe costs of misclassifyinga failed firmare likely
to be much greater than the costs associated with misclassifyinga
nonfailedfirm,it is importantto know the probabilityof misclassifying
a failed firm (Type I error) versus the probabilityof misclassifyinga
nonfailedfirm(Type II). (2) The difference in percentageerrorbetween
the ratios and the random-predictionmodel can vary substantially
by alteringthe probabilityof failure. If the probabilityof failure for
the sample differsfromthat of the total population, a comparisonof
percentageof total errorsis not verymeaningful.

The Type I and Type II errorsfor a given sample will be the same
as that forthe population,even if the probabilityof failureforthe pop-
ulation is differentfromthat of the sample.80The previous statement
requires one assumption-namely, that the ratio distributionsof the
sample accurately describe the ratio distributionsof the population.
Note also that the Type I and II errorsdiscussed are those obtained
whenthe cutoffpointswere selectedon the basis of minimizingthe total
numberof incorrectpredictionswithoutregardforthe Type I and Type
II errorsimplied. Since this mightbe a disastrousdecision rule where
the misclassificationcosts are asymmetrical,the cutoffpoints can be
easily alteredto any desiredratio of Type I to Type II errors.
One way of presentingType I and Type II errorsis throughcontingency
tables, as shownin Table 6.31 The row denotesthe predictionmade; the
columndenotesthe actual status of the firm;and each cell containsthe
numberof firmsfulfillingeach condition.For example,in the firstyear
beforefailure,the total numberof failed firmswas 79, of which62 were
correctlyclassifiedas failed and 17 were not. The total numberof non-
failed firmswas also 79, of which75 were correctlyclassifiedand 4 were
not. The Type I errorwas 17/79 or 22 per cent, and the Type II error
was 4/79 or 5 per cent. The total errorwas 21/158 or 13 per cent-the
numberpreviouslyreferred to as the percentageerror.
The Type I and Type II errorsare probabilitiesof errorconditional
upon the actual status of the firm.Anotherset of conditionalprobabili-
ties of erroris the probabilityof errorconditionalupon the prediction
made.The probability failure,is 4/66 or
of errorgiventhe prediction,
6 per cent.The probability
of errorgiventhe prediction,
nonfailure,is
17/72or 18 per cent.The two sets of conditionalprobabilitiesare merely
different ways of expressingthe same underlyingrelationships.Either
can be appropriatedependingupon the inferencesdesired.
It is interestingto comparethe Type I and II errorsof a ratio with
those obtained froma random-prediction model. For example, consider
a model constrainedso that the ratio of failed to nonfailedpredictions
is the same as that of the financialratio. In the firstyear beforefailure
the Type I and II errorswould be 58 per cent and 42 per cent,respec-
tively,whilein the fifthyear the Type I and II errorswould be 71 and 29
per cent,respectively.The random-prediction model could be altered in
many otherways, but the tenorof the inferenceswould remainthe same.
Analysis of Type I and Type II Errors

In the firstyear beforefailurethe cash-flowto total-debtratio has a
strikingability to classify both failed and nonfailed firmsto a much
0Unless otherwisestated, the paper shall referto the probability of a Type I error
simply as a Type I error,and similarlyfor a Type II error.
3 The results for the cash-flowto total-debt ratio are shown in Table 6. The results
for the other five ratios and for total assets appear in Table A-6 in the Appendix.

TABLE 6
Cash Flow to TotalDebt: Contingency
Table
Year One Year Two Year Three
Predicted ActualOutcome Predicted ActualOutcome Predicted ActualOutcome

Outcome Outcome Outcome
FLD NFLD Total FLD NFLD Total FLD NFLD Total
FLD 62 4 66 FLD 50 6 56 FLD 47 6 53
NFLD 17 75 92 NFLD 26 71 97 NFLD 28 69 97
Total 79 79 158 Total 76 77 153 Total 75 75 150
21 32 34
- = .13 3 .21 34-= .23
158 153 1 150
Year Four Year Five
PredictedOutcome ActualOutcome PredictedOutcome ActualOutcome
FLD NFLD Total FLD NFLD Total
FLD 33 2 35 FLD 31 3 34
NFLD 29 64 93 NFLD 23 60 83
Total 62 66 128 Total 54 63 117
26
31=.24 =22
greater extent than would be possible though random selection. The

Type I and Type II errorsare 22 per cent (vs. 58 per cent) and 5 per
cent (vs. 42 per cent). In the fifthyear before failure the errorsare
also smaller than would be obtained through random selection. The
Type I and Type II errorsare 42 per cent (vs. 71 per cent) and 4 per cent
(vs. 29 per cent). There is an extremelysmall probabilitythat random
predictioncould have done as well.32
The evidence furtherindicates that the ratio cannot classify failed
and nonfailedfirmswith equal success. In each year beforefailure,the
Type I erroris greaterthan the Type II error.In fact, the Type II
erroris remarkablystable over the five-yearperiod, while the Type I
errorincreasesas the timeperiodbeforefailureincreases.The differences
are 22 per centversus5 per cent (year one); 34 per centversus8 per cent
(year two); 36 per centversus8 per cent (year three); 47 per centversus
3 per cent (year four); and 42 per centversus4 per cent (year five). The
analysisoftheotherratioswouldhave thesame tenor,buttherelationships
32In bothyearstheprobability
is lessthanone in tenthousand.

wouldnotbe as strongas in thecase of the cash-flowratios.Both the Type

I and Type II errorswould tendto be higherfortheseratios.
The findingssuggestthat ratio analysis can be useful for at least five
years beforefailure,but ratios cannot be used indiscriminately.(1) Not
all ratios predictequally well. The cash-flowto total-debtratio predicts
quite well fiveyears beforefailure,but the same is not trueof the liquid-
asset ratios. (2) Ratios do not correctlypredict failed and nonfailed
firmswith the same degree of success. Nonfailed firmscan be correctly
classifiedto a greaterextentthan the failed firmscan. The implication
is important.Even with the use of ratios, investorswill not be able to
completelyeliminatethe possibilityof investingin a firmthat will fail.
This is a ratherunfortunatefact of life,since the costs in that event are
ratherhigh.
Limitationsof the ClassificationTest

One limitationof the test is that it treatsthe predictionsmade by the
ratio as dichotomous.Even if the decision does involve a dichotomous
choice, the conditionalerrorgiven the predictionmay differdepending
upon the magnitudeof the ratio. If the ratio is very far away fromthe
cutoffpoint,more confidencewill be placed in the predictionthan if it
were close. In otherwords,the magnitudeof the ratio can also provide
importantinformationregardingthe probabilityof error.This is not
revealed by a dichotomousclassificationtest, even when it is presented
in the formof contingency tables.
A second limitationis that the specificvalues of the cutoffpoints
obtained fromthe sample cannot be used in a decision-makingsituation.
One reason is that the cutoffpoints have been selected withoutregard
to the asymmetricalloss functionsof Type I and Type II errors.Another
is that the probabilityof failurefor the population is not the same as
that of the sample. Under these circumstancesit is unlikely that the
same cutoffpoint would be optimal.This does not implythat the conclu-
sions regardingthe predictiveability of ratios were incorrect,only that
the specificvalues of the cutoffpointscannot be directlycarriedover to
a decision-makingsituation.
Analysis of Likelihood Ratios

Financial ratios can be viewed as a way of assessingthe likelihoodof
failure.This approachpossessesneitherof the limitationsof the classifica-
tion test, and the ability to generalize about the predictiveability of
ratios is facilitated.In order to assess the likelihood ratios fromthe
financialratios, histogramsare prepared. Figure 2 contains histograms
for the cash-flowto total-debt ratio.33The horizontalaxis shows the
3Relative-frequencydistributionshave been prepared for the other five ratios and

RELATIVE
FREQUENCY
.35
.30
.25
.20 r
.15-
3 FAILED 5NONFAILED
.10 *
.05 I i
-.9 -.7 -.5 -.3 -.1 0 .1 .3 .5 .7 .9 1.1
CASH FLOW TO TOTAL DEBT(YEAR 1)
RELATIVE
FREQUENCY
.25
.20 r-
r- W E
.15 I _
.10 6 NONFAILED
3 FAILED
< ~~~~I ~ ~lI ~
.051-L J
n I Fn
-.9 -.7 .5 -.3 -.1 0 .1 .3 .5 .7 .9 1.1

- NONFAILED FIRMS - - - FAILED FIRMS
FIG. 2. Histogram.
valuesoftheratio,whiletheverticalaxisindicatestherelativefrequency
withwhichthe ratiosof failed(or nonfailed)firmsfall intoeach class
interval.For example,in the firstyear,28 per centof the ratiosof the
nonfailedfirmsfallin theinterval.1 to .2, and 21 per centof theratios
offailedfirms fallin theinterval-.1 to - .2.
Consistencyof Data
Histograms can provide insightsinto financial ratios, apart from a
For example, the inferencesdrawn from
likelihoodratio interpretation.
total assets.The data appearin Table A-7 in the Appendixand containthe same in-
formation thata histogramcontains.

RELATIVE
FREQUENCY
.30
.25
.20L
.15
.10 r -5 NON-FAILED
.05
-,n raw--
L_] m~?- r-iy
-.9 -.7 -.5 -.3 -.1 0 .1 .3 .5 .7 .9 1.1
RELATIVE
FREQUENCY
.40
l l
.35 I
.30 1
.25 I l
.20 I
l l
.15 - -
.10 lL. _ NONFAILED
.05 r- - I-
_ ~r-' II!1
-.9 .7 -.5 -.3 .1 0 .1 .3 .5 .7 .9 1
CASH FLOW TO TOTALDEBT(YEAR 4)
NONFAILED FIRMS - - - - FAILED FIRMS
FIG. 2-Continued
the profileand the classificationtest are also suggestedby an inspection

of the histograms.,
Figure 2 indicates that the distributionof nonfailedfirmsis remark-
ably stable in each of the years beforefailure.No shiftingover time is
apparentand, if the distributionsfromdifferent years were superimposed,
the overlap would be virtuallycomplete.In all instances,thereis a pro-
nouncedskewto the right,whilethe distributiondropsoffabruptlyon the
left.The stabilitysuggeststhatthe sampleofnonfailedfirmsis comparable
over time,even thoughtheirindustryand asset-size compositionchanges.
The distributionof failed firmsshiftsfartherto the left as failure ap-

RELATIVE
FREQUENCY
.35
.30 1
I I
.25 l I
I l
.20
I I
.15
_1 4 NONFAILED
.10.10 X II FAILED
.05 I I |
r- - - -L-- *_
-.9 -.7 -.5 -.3 -.1 0 -.1 .3 .5 .7 .9 1.1
CASH FLOW TO TOTAL DEBT (YEAR5)
NONFAILED FIRMS - - - FAILED FIRMS
FIG. 2-Continued
proaches,while the gap betweenthe failed and nonfailed firmsbecomes

greater.Similar behavior was noted in the comparisonof means in the
profileanalysis.
In the firstyear, the overlap of the two distributionsis small, which
leads to the low percentageerrorin the classificationtest. By the fifth
year,the degreeof overlap is greaterand is reflectedin the increasein the
percentageof error.The tail area of the failed firmsoverlaps onto the
nonfaileddistributionto a greaterextentthan the tail area of the non-
failed firmsoverlaps onto the failed distribution.The result is the dif-
ferencein Type I and Type II errorsmentionedearlier.Similarrelation-
ships exist in the other fiveratios but not to the same extentas in the
cash-flowto total-debtratio. The overlap is greater,an indicationthat
theirpredictivepoweris not as strong.
An inspectionof the histogramreinforcesthe conclusionsreached in
the earlier sections. However, the histogramsreveal additional aspects
of the ratios. Rather pronouncedskews were evident in most of the
distributions.The skews suggest that the data are not normally dis-
tributed.
Normalityof Data
The normalityof the data is best analyzed throughthe use of cumula-
tive density functions (cdf). The vertical axis indicates the fractile
associated with the correspondingvalue of the ratio, which is shown
on the horizontalaxis. The fractileis the proportionof the distribution
whose values are equal to or below a given value of the ratio. The
fractileforthe ithobservationis the "i/n + 1" fractile,where "n" is the
numberof failed (or nonfailed) firms.The ith observationis determined

FRACTILE
99
,98 x + 5 NONFAILED
x +
.90 9 XXX XX + X OOO ODENOTESFIRMS+>? 00 00

NONFAILED
+COO
.80- ++f+ 08 0,0000
.01 ++ 0
.60-3.
.50-
.40- + ? + FAILEDFIRMS(YEAR1
XDENOTES )1
.30 X 04
X Xxx) ++++
.10 x X++
.05- X X ++ + ++ oDENOTES NONFAILEDFIRMS-(YEAR1)
<-3 ~~~~~~+0 0 XDENOTES FAILED FIRMS (YEAR 1)
.024-3FAILED (YR. 1) + DENOTES FAILED FIRMS (YEAR 5)
.01 +
<-1 FAILED (YR. 5)
-.9 -.7 -. -.3 -.1 0 .1 .3 .5 .7 .9 1.1

CASH FLOW TO TOTALDEBT (YEAR 1)
FIG. 3. CDF.
by arraying the data in ascending order. The lowest value is the first
observation, while the highest value is the nth.The .70 fractile of the failed
firms(Fig. 3) in the firstyear is zero, while the .70 fractileforthe non-
failed firmsis .40. The vertical axis is calibrated accordingto the arith-
metic probabilityscale. If the data are normal, the distributionwill
appear as a straightline. Departure fromlinearitycan be interpretedas
departurefromnormality.
The cash-flowto total-debtratio has pronouncedskews in each of the
years before failure.The other five ratios also do not conformto the
normalityassumption.4 Simple transformations(e.g., log and square
root) have been performedon the ratios in the hope that the transformed
variables would be approximatelynormal. The initial results indicate
that the transformed variables are almost as badly skewedas the original
variables. The departure fromnormalityhas serious implications for
attemptsto derive a multiratiomodel, since most of the existingmulti-
variate techniquesrely upon the normalityassumption.
Likelihood Ratios
None of the discussionthus far has been directedtoward the use of
ratios as assessmentsof likelihoodratios. The study of financialratios
as predictorsof failure is placed in its broadest context throughthe
discussion of likelihoodratios. Their use is essentially a Bayesian ap-
proach and can be explained most convenientlyby referringto the
I The data in Table A-7 can be used to formcdf'sforthe otherfiveratios.Table
A-7 appearsin theAppendix.

followingtableau:
Event PriorProbability Likelihood JointProbability PosteriorProbability
Fail P(F) P(R/F) P(R,F) P(F/R)

Not fail P(F) P(R/F) P(R,P) P(P/R)
1.00 P(R) 1.0
The problemof predictingfailurecan be viewed as a problemin as-

sessingthe probabilityof failureconditionalupon the value of the ratio-
P(F/R). The conditional probability of failure is shown in the last
column,whichis headed posteriorprobability.In arrivingat estimatesof
the conditionalprobabilityof failure,the possible events are viewed as
being dichotomous-eitherthe firmwill fail or it will not. Before look-
ing at the financial ratios of the firm,certain prior probabilities are
formed.The prior probabilities-P(F) and P() -may be based upon
several factors,such as the unconditionalprobabilityof failure for all
firms,industry,asset size, or quality of management.Since failure and
nonfailureare the only two events that can occur, the sum of prior
probabilitiesis 1.00.
Afterthe financialratio is observed,assessmentsof the likelihoodof
failureand nonfailureare formed.The likelihoodof failureis the proba-
bilitythat the observednumericalvalue of the ratio would appear if the
firmwere failed-P(R/F); the likelihoodof nonfailureis the probability
that the specificvalue of the ratio would be observed if the firmwere
nonfailed-P(R/F). The joint probabilitiesare the productof the prior
probabilitiestimesthe likelihoodestimates.35The sum of the joint proba-
bilities is the marginalprobability-P(R)-the probabilitythat a ratio
of the observednumericalvalue could occur.
The posteriorprobabilityis the quotient of the joint probabilitydi-
vided by the marginalprobability.36 The sum of the posteriorprobabili-
a The calculationis
P(F)P(R/F) = P(R,F)
P (P) P (R/P) -P (RF)
and is based uponBayes' Theoremwhichstates: GiventwoeventsA and B,
P(AB =
(A/B).
P(B)
38The calculation is
P(RF) = P(F/R)
P(R?)
P(RF) P(F/R)
and is also based uponBayes' Theorem.

ties must be 1.00. The posteriorprobabilityis the probabilityof failure

(or nonfailure)afterthe ratio analysis.
The tableau can also be expressedin termsof odds ratios ratherthan
probabilities.For example,if the probabilityan event will occur is .75
(which implies the probabilityof nonoccurrenceis .25), it can also be
said that the odds are 3 to 1 (i.e., .75/.25) that the event will occur.
In fact,in many cases it is commonpracticeto state the relationshipsin
termsof odds ratherthan probabilities-the predictionof the outcomeof
horseraces and boxingmatchesis a good example.The priorprobabilities
are replaced by the prior-oddsratio, the likelihood estimates by the
likelihood-oddsratio, and the posteriorprobabilitiesby the posterior-
odds ratio.37 The followingrelationshipexists among the three odds
ratios:38
(Prior-oddsratio) x (likelihood-oddsratio) = (posterior-oddsratio).

The discussion could be conducted in terms of the posterior-odds
ratio. However,the ratio would largelybe affectedby the probabilityof
failure for this particular sample (i.e., .50), which is vastly different
fromthe probabilityfor all firmsin the economy (i.e., less than .01).
The likelihoodratios are unaffectedby the probabilityof failure and
thereforecarrywith them a degreeof generality.For example, consider
a sample whose financial-ratiodistributionsaccuratelyreflectthe finan-
cial-ratio distributionsof the population. The numericalvalues of the
likelihood ratios, which are derived from the financial ratios of the
sample, are the same ones that would apply to the population, even
thoughthe frequencyof failure in the sample is vastly differentfrom
that of the entirepopulation.
If the likelihood-oddsratio in favor of failureis greaterthan 1 (one),
the user of the ratio, after having looked at the firm'sratio, will feel
that the firmis more likely to fail-the higherthe likelihoodratio, the
strongerthe feeling.If the likelihoodratio is less than 1, the user of
the ratio will feel that the firmis less likelyto fail-the lowerthe ratio,
the strongerthe feeling.If the likelihood ratio is exactly 1, the prior
feelingsof the user are unchangedafterlooking at the ratio-the poste-
The odds are definedas follows:
Prior-odds ratio P(F)
P(P/)
Likelihood-odds ratio P(R/F)
Posterior-odds ratio P(F/R)
38 PP(F) P(R/F) P (R,F) P(R,F) P(R) P (F/R)

P(P) P(R/IP) P(R,P) P(R) P(R,P) P(P/R)

rior-oddsratio will be numericallyequal to the prior-oddsratio. The

role of financialratios is to provide assessmentsof the likelihoodratio.
The informationcontent of the ratios can be evaluated in terms of
the degreeto whichtheychangethe priorfeeling.
The discussionis based upon estimatesderived fromthe histograms.
The likelihoodestimates-P(R/F) and P(R/1) -are obtainedby meas-
uring the heightsof the failed and nonfailed distributionsat a given
value of the ratio. The likelihoodratio is the ratio of those two heights.
For example,in the firstyear beforefailure,for the cash-flowto total-
debt ratio, the likelihoodestimatesforthe ratio interval .0 to .1 are .18
(failed firms) and .11 (nonfailedfirms).The likelihoodratio would be
1.64 (.18/.11). In the interval -.1 to .0, the likelihood ratio would be
16.00, while in the interval -.2 to -.1, the likelihoodratio is undefined
since thereare not any nonfailedfirmsin that interval.Presumably,the
likelihood ratio would be extremelyhigh, judging by the behavior of
the distributions in the surroundingclass intervals.
If the ratio of the heightsof the distributionsis blindlyused to com-
pute the likelihoodratio, some rathererraticnumberswill be obtained.
A more reasonable approach would be to smooth the data by means
of a movingaverage technique or an exponentialfunction.No attempt
was made to smooththe data nor werethe numericalvalues of the likeli-
hood ratio computed.The discussion shall be in terms of the general
behaviorof likelihoodratios ratherthan theirspecificnumericalvalues.
The data are available forthe readerto carryout the numericalcompu-
tations if he so desires.
Analysis of Evidence (Likelihood Ratios)

In the firstyear beforefailure,the likelihoodratio in favor of failure
is eithervery large or very small over most of the range of the ratio.
Looking at the financial ratio will lead the user to change his prior
impressionsa great deal. Even five years before failure,the likelihood
ratio still takes on high values. The implicationis that the ratio can
conveyusefulinformationin determiningsolvencyforat least fiveyears
prior to failure. In the tail areas of the distributions,the likelihood
to quantifybecause of the scarcenessof observations
ratios are difficult
in these areas. The developmentof smoothingtechniquesand an increas-
ing of the sample size would produce more accurate measures of the
likelihoodratios.
In the firstyear, the relationshipbetweenthe likelihoodratio and the
financial ratio is monotonic-the higherthe financial ratio, the lower
the likelihoodratio. This simplerelationshipmakes the interpretation of
the financialratio relativelystraightforward.However, in the fifthyear
the likelihoodratio decreases as the financialratio increases over most,
but not all, values of the ratio. In the highestvalues of the ratio, the

FINANCIAL RATIOS AS PREDICTORS OF FALURE 99
likelihoodratio increases slightly.The implicationis that it is slightly

more risky for a firmto have a very high cash-flowto total-debtratio
than to have a lowerone in a rangewherethe bulk of the nonfailed firms
appear. Here the interpretation of the ratio is not quite so simple as it
was in the firstyear. The U-shaped behavior of the likelihoodratio was
also observed in the other five ratios, even in the firstyear before
failure.The phenomenondeservesmore attention,and its explanationis
one of the areas plannedforfutureresearch.
The ClassificationTest Revisited

Since the probabilityof failureforthe sample was .50 and the costs of
misclassificationwere implicitlytreated as being symmetrical,the opti-
mal cutoffpoint was located where the likelihoodratio equaled 1.00. If
either of these conditionsis altered, the cutoffpoint will have to be
shiftedto anothervalue of the likelihoodratio,whichcan be determined
once the costs of misclassificationand the probabilityof failure have
been specified.39When the likelihood-ratioapproach is used, the shifting
of the cutoffpoint involves no substantivechange in the analysis. The
same likelihoodratiosstillapply,subjectof courseto samplingerror.
Once a classificationhas been made, the likelihoodratios can be used
to assess the conditionalprobabilityof error,which takes into account
how farthe financialratio is fromthe cutoffpoint.This errorassessment
is superiorto the conditionalerrorderived fromthe contingencytable,
because the latter conditionalerrorignoreshow far a given firm'sratio
may be fromthe cutoffpoint.
CONCLUDING REMARKS
Implication for Accounting

In a very real sense, the title of this paper should not be "Financial
Ratios as Predictorsof Failure" but rather"AccountingData as Predic-
tors of Failure." My primaryconcernis not with ratios per se but with
the accountingdata that comprisethe ratios. One premiseis crucial to
an acceptance of this viewpoint.The premise is that accountingdata
can be evaluated in termsof theirutilityand that utilitycan be defined
in termsof predictiveability. Possibly the most importantcontribution
to be made by this paper is to suggesta methodologyforthe evaluation
of accountingdata for any purpose,not merelyfor solvencydetermina-
tion. I feel this approach will be helpfulin resolvingmany of account-
ing's controversies.
The financial-leasecontroversycould be subjected to tests similar to
the ones used here. The efficacyof capitalizing financial leases could
89A more complete discussion of this topic appears in Professor Neter's comments
and in my reply to his comments.

be evaluated by computingtwo sets of financialratios. One set would

include the capitalized value of leases as debt, while the other set
would not. The set of ratios that best predicted failure would suggest
whetheror not financial leases should be capitalized. A similar ap-
proach could be used for other controversiesand for other predictive
purposes.40
Suggestionsfor Future Research

The analysis conductedhere has been a univariate analysis-that is,
it has examinedthe predictiveability of ratios,one at a time.It is possi-
ble that a multiratioanalysis,using several differentratios and/orrates
of change in ratios over time,would predicteven betterthan the single
ratios. Some preliminaryeffortshave been undertakento develop multi-
ratio models,but the resultshave not been very encouragingin the sense
that the best singleratio appears to predictabout as well as the multi-
ratiomodels.
As stated in the introductoryremarks,the immediatepurpose of this
study was not to findthe best predictorof failurebut ratherto investi-
gate the predictiveability of financial ratios. That evaluation is en-
hanced by a comparisonwith otherpredictorsof failure.Initially,efforts
will be directedtoward investigatingmarket rates of return (i.e., divi-
dends plus price changes) as predictorsof failure.Market rates of return
will providea verypowerfultest sincetheyreflectall of the sourcesof in-
formationavailable to investors-sourcesthat not onlyincludeaccounting
data but otherkindsof information as well.
A replicationon a populationof smallerfirmswould be anotheravenue
for additional research.The Merwin study was based upon a sample of
firmswhose average asset size was $50,000,and yet the studyproduced
differences in means comparable to those observed here. This suggests
that the replicationon smaller firmswould be worthwhile,if the data
were accessible.
The profileanalysis indicatedthat the mean currentratio of the failed
firmswas above the magic "2:1" standard in all five years. In fact,in
the finalyear the mean value is 2.02-which is about as close to 2.00 as
possible. The evidence hints that failed firmsmay attemptto window
dress. However,when the failed firmsare comparedwith the nonfailed,
their weaker position is evident. Perhaps the nonfailed firmswindow
dressmoresuccessfullythan do the failed firms.If so, attemptsto window
dressmay tend to improvethe predictivepowerof ratiosratherthan im-
pair it, as is oftensuggested.No pretenseis made that this is anything
otherthan a causal glance at the data. However, it does suggestan in-
triguinghypothesisto be investigatedat some futuredate.
40Anexamplewould be Phillip Brown's"The PredictiveAbilitiesof Alternative
Income Concepts"(unpublishedmanuscriptpresentedbeforethe Workshopin Ac-
countingResearch,University
of Chicago,GraduateSchoolof Business,April,1966).

Summaryand Conclusions
Beforerestatingthe conclusions,I would like to provideone finalword
of warningregardingthe uncertaintysurroundingthis empiricalstudy.
There are many factorsthat have preventeda measurementof the "true"
predictiveability of ratios. In addition to the factors previouslymen-
tioned,two more come to mind. Both stem fromthe fact that financial
ratiosare in widespreaduse as measuresof solvency.
If ratios are used to detectthe financial"illness" of a firm,theremay
be many firmswhose illnesses were detectedbeforefailureoccurred.In
these cases, the propertreatmentwas applied, and the firmsdid not fail.
The sample of failed firmswill include those firmswhose illnesses were
not detectablethroughratios. This is a biased sample for investigating
the usefulnessof ratios. An importantpiece of informationis missing-
how many firmswere saved fromfailurebecause theirproblemswere de-
tectedin timethroughthe use of ratios? Such information would be diffi-
cult, if not impossible,to obtain. However,this argumentsuggeststhat
the studymay be understating the usefulnessof ratios.
On the other hand, ratios are used by financialinstitutionsto deter-
minethe credit-worthiness of its borrowers.In many cases, lines of credit
may be severed by the lendinginstitutionsbecause the borrowingfirm
failed to improveits financialratio to a "respectable"level. This argu-
ment suggeststhat the predictiveability of ratios may be overstated,
relativeto what it would be if ratios were not so popular. There exists a
countlessnumberof argumentson both sides, regardingthe possible bi-
ases in the data. This briefdiscussionshould suggestthat the conclusions
that followare arrivedat withmuchless than perfectcertainty.
The ratio distributionsof nonfailedfirmsare quite stable throughout
the fiveyears beforefailure.The ratio distributionsof the failed firms
exhibita markeddeteriorationas failureapproaches.The resultis a wid-
eninggap betweenthe failed and nonfailedfirms.The gap producesper-
sistentdifferences in the mean ratios of failed and nonfailedfirms,and
the difference increasesas failureapproaches.
For the cash-flowto total-debtratiothe overlapof the two distributions
is small in the firstyear beforefailure.By the fifthyear,the overlap has
increasedbut the two distributionsare still distinguishable.The degree
of overlap is reflectedin the abilityto correctlypredictthe failurestatus
of firms.The cash-flowto total-debtratio has the ability to correctly
classifyboth failed and nonfailedfirmsto a much greaterextentthan
would be possible throughrandom prediction.This ability exists for at
least fiveyears beforefailure.
Althoughratio analysis may provide useful information,ratios must
be used withdiscretion:(1) Not all ratiospredictequally well. The cash-
flowto total-debt ratio has excellentdiscriminatorypower throughout
the five-yearperiod. However, the predictivepower of the liquid asset
ratios is much weaker. (2) The ratios do not predict failed and non-

failed firmswiththe same degreeof success. Nonfailed firmscan be cor-

rectlyclassifiedto a greaterextentthan can failed firms.The implication
is thatthe investorwill not be able to completelyeliminatethe possibility
of investingin a firmthat will fail. This is a ratherunfortunatefact of
life,sincethe costsin that eventare ratherhigh.
When financialratios are used to assess likelihoodratios,the cash-flow
to total-debtratio produceslarge likelihoodratios,even fiveyears before
failure.Looking at the financialratio will lead the user to shifthis prior
feelingsa great deal. In general,the relationshipbetweenthe likelihood
ratio and the cash-flowto total-debtratio is a monotonicone. However,
in the fifthyear beforefailure,thereis a slightincreasein the likelihood
ratio in the highvalues of the cash-flowratio. This findingimpliesthat it
is more risky to have a very high ratio than to have a lower one in a
rangewheremostof the nonfailedfirmsappear.
The data exhibita remarkabledegreeof consistencyamongthemselves
and with previousstudies. Subject to the reservationsthat must accom-
pany inferencesdrawn fromthis study,the evidenceindicatesthat ratio
analysis can be useful in the predictionof failureforat least fiveyears
beforefailure.

APPENDIX
TABLE A-I
Year of Failure
Year of Failure No. of Failed Firms
1954 2
1955 6
1956 14
1957 7
1958 6
1959 4
1960 7
1961 10
1962 9
1963 10
1964 4
Total 79
103

TABLE A-2
IndustrialClassification
No. of
Industrya Failed
Firms
Manufacturing
Food products(201) 1
Grainmill products(204) 1
Bakeryproducts(205) 3
Confectionery (207) 1
Beverages (208) 3
Textile (221) 4
Women'sapparel (233) 1
Industrialchemicalsand plastics (281) 3
Petroleumrefining(291) 3
Rubber products(301) 1
Leather products(311) 2
Clay products(325) 1
Concrete,gypsum(327) 1
Iron and Steel foundries(331) 2
Nonferrousmetals (333) 1
Cutlery,etc. (342) 3
Heating and plumbing(343) 2
Structuralmetal products(344) 1
Etching,engraving(347) 2
Engines,industrialmachinery(351) 3
Service,industrialmachines(358) 1
Householdappliances,etc. (363) 5
Commercialequipment(366) 1
Electronics(367) 6
Motor vehicles (371) 5
Aircraft(372) 3
Scientificinstruments(381) 1
Watches (387) 2
Toys, sportinggoods (394) 1
Nonmanufacturing
Public warehouse(422) 1
Hardware,wholesale (507) 1
Generalmerchandise(532) 3
Grocerystores (541) 2
Apparelstores (561) 3
Furniturestores (571) 1
Drug stores (591) 2
Personal services(721) 1
Motion pictures(781) 1
a SIC number.
The numberin parenthesesindicatesthe three-digit

TABLE A-3
Number of Observations on Which Financial-Statement Data Could Be Obtained
Year beforeFailure
Item
. 2 3 4 5
Failed firms.79 76 75 62 54
Nonfailed firms.79 77 75 66 63
Total number.158 153 150 128 117

TABLE A-4
Percentage Error for 30 Ratios on Dichotomous Classification Test"
Year beforeFailure
Ratio
12 3 4 5
Group I
Cash flow .14 .20 .29 .37 .44
Sales (.11) (.19) (.24) (.33) (.31)
Cash flow .10 .20 .24 .28 .28
Total assets (. 10) (.17) (.20) (.26) (.25)
Cash flow .13 .21 .28 .38 .37
Net worth (. 11) (.16) (.24) (.31) (.32)
Cash flow .13 .21 .23 .24 .22
Total debt (.10) (.18) (.21) (.24) (.22)
Group11
Net income .13 .22 .28 .35 .31
Sales (. 09) (.16) (.24) (.28) (.27)
Net income .13 .21 .23 .29 .28
Total assets (.12) (.15) (.22) (.28) (.25)
Net income .13 .26 .26 .34 .40
Net worth (.10) (.18) (.26) (.31) (.28)
Net income .15 .20 .22 .26 .32
Total debt (.08) (.16) (.20) (.26) (.26)
Group III
Currentliabilities .30 .41 .36 .40 .46
Total assets (.27) (.28) (.34) (.33) (.30)
Long-termliabilities .36 .45 .42 .47 .51
Total assets (.32) (.40) (.40) (.38) (.41)
Current+ long termliabilities .23 .30 .36 .39 .38
Total assets (.19) (.26) (.29) (.31) (.33)
Current+ longterm+ preferred .19 .25 .34 .27 .28
Total assets (.19) (.24) (.28) (.24) (.27)
Group IV
Cash .28 .29 .30 .36 .38
Total assets (.25) (.28) (.30) (.34) (.31)
Quick assets .38 .42 .36 .48 .40
Total assets (.34) (.36) (.36) (.37) (.34)
a Top rowofnumbersaretheresultsofthesecondtest;thebottomrowofnumbers,
enclosedin parentheses,referto the resultsof the firsttest.

TABLE A - 4- Continued
Year before Failure
Ratio
1 2 3 4 5
Group IV-Continued
Currentassets .38 .48 .48 .47 .49
Total assets (.37) (.44) (.43) (.43) (.38)
Total assets (.20) (.30) (.33) (.35) (.35)
GroupV
Cash .22 .28 .36 .38 .38
Currentliabilities (.22) (.24) (.28) (.34) (.29)
Quick assets .24 .32 .40 .34 .37
Currentassets .20 .32 .36 .38 .45
GroupVI
Cash .34 .24 .36 .43 .45
Sales (.30) (.24) (.34) (.39) (.41)
Receivables .46 .45 .46 .43 .42
Sales (.40) (.39) (.38) (.37) (.38)
Inventory .47 .50 .54 .48 .53
Sales (.40) (.50) (.47) (.42) (.42)
Quick assets .46 .47 .45 .52 .44
Sales (.40) (.42) (.40) (.48) (.42)
Cur. assets .44 .51 .48 .49 .51
Sales (.42) (.40) (.42) (.47) (.47)
Sales (.23) (.27) (.35) (.40) (.37)
Net worth .32 .36 .44 .42 .40
Sales (.28) (.33) (.38) (.38) (.38)
Total assets .37 .42 .38 .42 .44
Sales ( .34) (.40) (.34) (.38) (.39)
CashInterval ~.33 .27 .35 .42 .43

Cash Interval (.26) (.27) (.32) (.42) (.38)
Defensive
Interval39 .41 .46 .48 .51
DefensiveInterval (.34) (.40) (.41) (.47) (.43)
No-creditInterval.23 .38 .43 .38 .37

No-creditInterval | (.23) (.31) (.30) (.35) (.30)
^ Top rowofnumbers aretheresultsofthesecondtest; thebottomrowofnumbers,
enclosedin parentheses,referto the resultsofthe firsttest.
107

TABLE A-5
CutoffPoints Used in Classification Test for Six Ratiosa
Year before Failure
Ratio 1 2 3 4 5
A B A B A B A B A B
Cash flow
.03 .07 .05 .07 .10 .09 .09 .09 .11 .11
Total debt
Net income
.00 .02 .01 .02 .03 .03 .02 .02 .04 .03
Total assets
Total debt
..57 .57 .51 .49 .53 .50 .58 .57 .57 .57
Total assets
Working capital .19 .27 .33 .28 .26 .26 .40 .24 .43 .29
Total assets
Current assets 1.6 1.6 2.3 1.7 2.3 1.8 2.6 1.8 2.8 2.1
Currentliab.
No credit interval -.04 -.04 .03 -.02 .01 -.01 .00 -.01 .04 -.02
Total assetsb 4.6 4.6 5.2 5.2 4.4 4.4 5.5 4.0 4.8 2.5
A and B columns referto cutoffpoints used on the two subsamples, A and B.

b Expressed in millions of dollars.

TABLE A-6
ContingencyTables: Five Ratios and Total Assets
ActualOutcome
PredictedOutcome Net Incometo Total Debt to WorkingCapital to

Total Assets Total Assets Total Assets
Fa F T F F T F P T
Year One
F 66 9 75 56 7 63 54 13 67
F 13 70 83 23 72 95 25 66 91
T 79 79 158 79 79 158 79 79 158
Year Two
F 57 13 70 53 16 69 48 24 72
F 19 64 83 23 61 84 28 53 81
T 76 77 153 76 77 153 76 77 153
Year Three
F 49 9 58 36 12 48 38 13 51
F 26 66 92 39 63 102 37 62 99
T 75 75 150 75 75 150 75 75 150
Year Four
F 34 9 43 43 14 57 30 25 55
F 28 57 85 19 52 71 32 41 73
T 62 66 128 62 66 128 62 66 128
Year Five
F 36 15 51 25 4 29 36 30 66
F 18 48 66 29 59 88 18 33 51
T 54 63 117 54 63 117 54 63 117
ActualOutcome
PredictedOutcome CurrentRatio No-CreditInterval Total Assets
F P T F F T F P T
Year One
F 55 7 62 52 9 61 46 27 73
F 24 72 96 27 70 97 33 52 85
T 79 79 158 79 79 158 79 79 158
Year Two
F 44 18 62 49 31 80 46 34 80
F 32 59 91 27 46 73 30 43 73
T 76 77 153 76 77 153 76 77 153
Year Three
F 42 21 63 46 36 82 40 32 72
F 33 54 87 29 39 68 35 43 78
T 75 75 150 75 75 150 75 75 150
Year Four
F 34 21 55 35 21 56 31 32 63
F 28 45 73 27 45 72 31 34 65
T 62 66 128 62 66 128 62 66 128
Year Five
F 33 31 64 34 23 57 25 26 51
F 21 32 53 20 40 60 29 37 66
T 54 63 117 54 63 117 54 63 117
aF denotesfailed,P denotesnonfailed,and T denotestotal. Cells containnumber
conditionsregardingpredictedand actual outcomes.
offirmsfulfilling
109

TABLE A -7
Relative Frequency Distribution of Five Ratios and Total Assets
Year before Failure Year before Failure
Class Interval of Net Class Interval ot Total

Income to Total Assets Debt to Total Assetsb 5
1
Fa | F F F F
Below -.65 .051 .000 .000 .000 0 to .1 .013 .063 .018 .032
-.65 to -.60 .025 .000 .000 .000 .1 to .2 .025 .152 .093 .206
-.60 to -.55 .000 .000 .000 .000 .2 to .3 .038 .152 .112 .143
-.55 to -.50 .000 .000 .000 .000 .3 to .4 .089 .240 .112 .143
-.50 to -.45 .038 .000 .018 .000 .4 to .5 .076 .165 .093 .221
-.45 to -.40 .013 .000 .000 .000 .5 to .6 .102 .152 .183 .201
-.40 to -.35 .013 .000 .000 .000 .6 to .7 .152 .051 .183 .015
-.35 to -.30 .051 .000 .018 .000 .7 to .8 .038 .025 .074 .032
-.30 to -.25 .038 .000 .018 .000 .8 to .9 .076 .000 .074 .000
-.25 to -.20 .089 .000 .018 .000 .9 to 1.0 .102 .000 .037 .000
-.20 to -.15 .127 .000 .037 .000 1.0 to 1.1 .063 .000 .000 .000
-.15 to -.10 .076 .000 .037 .000 1.1 to 1.2 .038 .000 .000 .000
-.10 to -.05 .102 .000 .037 .016 1.2 to 1.3 .076 .000 .000 .000
-.05 to -.00 .202 .038 .093 .032 1.3 to 1.4 .025 .000 .000 .000
.00 to .05 .152 .367 .425 .318 1.4 to 1.5 .013 .000 .018 .000
.05 to .10 .025 .418 .223 .430 1.5 to 1.6 .025 .000 .000 .000
.1O to .15 .000 .140 .037 .127 1.6 to 1.7 .000 .000 .000 .000
.15 to .20 .000 .025 .000 .048 1.7 to 1.8 .000 .000 .000 .000
.20 to .25 .000 .013 .018 .032 1.8 to 1.9 .000 .000 .000 .000
.25 to .30 .000 .000 .018 .000 1.9 to 2.0 .000 .000 .000 .000
Above .30 .000 .000 .000 .000 above 2.0 .038 .000 .000 .000
Total Relative 1.0001.0001.0001.000 Total Relative 1.0001.0001.0001.000

Frequency Frequency
Year before Failure Year before Failure
Class Interval of Class Interval of__ -____ ___

Working Capital to 1 Class Interval of 5
Total AssetsCurnRai15
F P p P F
Below -1.0 .025 .000 .000 .000 Oto .5 .139 .000 .018 .000
-1.0 to -.9 .013 .000 .000 .000 .5 to 1.0 .202 .000 .055 .000
-.9 to -.8 .000 .000 .000 .000 1.0 to 1.5 .254 .013 .223 .079
-.8 to -.7 .000 .000 .000 .000 1.5 to 2.0 .165 .152 .202 .143
-.7 to -.6 .013 .000 .000 .000 2.0 to 2.5 .038 .254 .186 .159
-.6 to -.5 .025 .000 .000 .000 2.5 to 3.0 .038 .139 .129 .190
-.5 to -.4 .025 .000 .000 .000 3.0 to 3.5 .038 .152 .055 .095
-.4 to -.3 .051 .000 .000 .000 3.5 to 4.0 .013 .063 .037 .048
-.3 to -.2 .038 .000 .000 .000 4.0 to 4.5 .038 .038 .000 .048
-.2 to -.1 .038 .000 .037 .000 4.5 to 5.0 .013 .038 .055 .048
-.1 to 0 .114 .000 .037 .000 5.0 to 5.5 .000 .025 .000 .037
o to .1 .139 .025 .148 .048 5.5 to 6.0 .000 .013 .000 .048
.1 to .2 .177 .089 .130 .095 6.0 to 6.5 .000 .051 .000 .037
.2 to .3 .114 .152 .148 .111 6.5 to 7.0 .000 .013 .000 .048
.3 to .4 .063 .254 .183 .238 7.0 to 7.5 .000 .025 .000 .000
.4 to .5 .063 .228 .148 .206 7.5 to 8.0 .000 .000 .000 .000
.5 to .6 .038 .127 .111 .190 8.0 to 8.5 .038 .013 .018 .000
.6 to .7 .051 .114 .018 .048 8.5 to 9.0 .000 .000 .000 .000
.7 to .8 .013 .013 .037 .064 9.0 to 9.5 .013 .000 .000 .000
.8 to .9 .000 .000 .000 .000 9.5 to 10.0 .013 .000 .000 .000
.9 to 1.0 .000 .000 .000 .000 Abovel0.0 .000 .013 .018 .018
Total Relative 1.0001.0001.0001.000 Total Relative 1.0001.0001.000 1.000

Frequency Frequency
110

TA B LE A - 7-Continued
Year beforeFailure Year beforeFailure
Class IntervalofNo- | Class Intervalof

CreditInterval 1 Total Assetsb 1
F F F l F P F p
Below -.9 .013 .000 .037 .000 0 to 1 .051 .000 .074 .016
-.9 to -.8 .013 .000 .000 .000 1 to 2 .127 .051 .185 .111
-.8 to -.7 .013 .000 .000 .000 2 to 3 .139 .114 .111 .175
-.7 to -.6 .013 .000 .000 .000 3 to 4 .114 .127 .148 .175
-.6 to -.5 .025 .000 .000 .000 4 to 5 .238 .114 .130 .048
-.5 to -.4 .051 .000 .000 .000 5 to 6 .051 .114 .074 .064
-.4 to - .3 .102 .000 .018 .000 6 to 7 .000 .051 .055 .032
-.3 to - .2 .076 .013 .055 .000 7 to 8 .089 .063 .074 .048
-.2 to -.1 .165 .000 .129 .048 8 to 9 .038 .038 .018 .095
-.1 to 0 .263 .316 .315 .286 9 to 10 .025 .051 .037 .079
0 to .1 .190 .392 .260 .334 10 to 11 .051 .089 .018 .032
.1 to .2 .063 .165 .074 .206 11 to 12 .000 .025 .000 .016
.2 to .3 .000 .063 .055 .032 12 to 13 .000 .000 .000 .000
.3 to .4 .013 .000 .018 .032 13 to 14 .013 .025 .000 .016
.4 to .5 .000 .000 .000 .032 14 to 15 .013 .013 .000 .032
.5 to .6 .000 .025 .018 .000 15 to 16 .013 .013 .018 .016
.6 to .7 .000 .000 .000 .000 16 to 17 .000 .038 .000 .016
.7 to .8 .000 .013 .000 .000 17 to 18 .013 .013 .018 .000
.8 to .9 .000 .00 .000 .000 18 to 19 .000 .013 .000 .016
.9 to 1.0 .000 .000 .000 .000 19 to 20 .000 .000 .018 .000
1.0 to 1.1 .000 .000 .018 .000 Above2O .038 .051 .018 .016
Above 1.1 .000 .013 .000 .032
Total Relative 1. 000 1. 000 1. 000 1. 000 Total Relative 1. 000 1. 000 1. 000 1. 000
Frequency Frequency
a F denotes distribution of failed firms,while P denotes distribution of nonfailed
firms.
b In millions of dollars.


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

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

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

Uploaded by

Copyright:

Available Formats

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

Financial Ratios As Predictors of Failure

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

At the turnof the century,ratio analysis was in its embryonicstate. It

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

bers consistof financialstatementitems.A thirdterm,predictiveability,

Selection of Failed Firms

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

industrialfirms-investedcapital representingthe contributionsof sev-

Selection of Nonfailed Firms

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

features: (1) the firmswere grouped accordingto industry (using the

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

10:W. J. Baumol, "The Transactions Demand for Cash: An Inventory Theoretic

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

analysis, which does permit inferencesregardinga single observation.

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

two conditions:(1) it mustbe closestin asset size to the failed firm-spe-

Collection of Financial Statement Data

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

GROUP I (CASH-FLOW RATIOS) GROUP V (LIQUID-ASSET TO CUR-

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

will becomethose mostmanipulatedby management(an activityknown

Comparisonof Mean Values

Theory of Ratio Analysis

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

flow) model8 The purposeof introducingthe cash-flowmodel is not to

praisingthe DefensivePositionof the Firm: The IntervalMeasure,"The Accounting

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

Cash flowto total debtb Nonfailed > failed

Comparisonof Mean AssetSize

'Paul J. Fitz Patrick,"A Comparisonof Ratios of SuccessfulIndustrialEnter-

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

*CASH FLOW NET INCOME TOTAL DEBT

+.05 -A.1 .51 b

.42- 3.5 . +.15

.30 -o c 3.0- +.05

.11 / 2.5- -.05-

LEGEND: .--NONFAILED FIRMS - - - FAILED FIRMS

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

firmswith those of discontinuedfirmsfor the period 1926 to 1936.23 A

search, Bulletin No. 15 [Urbana, Ill.: University of Illinois Press, 1927]).

for an industry) may have serious limitations.

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

For example,a bank's lendingdecision can be viewed as a dichotomous

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

Cash flow .13 .21 .23 .24 .22

Net income .13 .20 .23 .29 .28

Total debt .19 .25 .34 .27 .28

Workingcapital .24 .34 .33 .45 .41

.20 .32 .36 .38 .45

Total Assets .38 .42 .45 .49 .47

the resultsof the secondtest. The bottomrowrefersto

fromthe othersubsample.The second test is similarto the classification

Analysis of Evidence (Percentage Error)

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

withthe bestratio.The total-debtto total-assetsratio predictednextbest,

Implications of Imperfect Pairing for Asset Size

This content downloaded on Wed, 12 Dec 2012 16:20:50 PM

Cash flowto total assetsa

Difference .00 .07 .04 .05 .00

Difference .02 .06 .07 (.09) (.02)

Difference .04 (.02) .04 .00 (.11)