A Discriminant Analysis of Predictors of Business Failure Edward B. Deakin Journal of Accounting Research, Vol. 10, No. 1 (Spring, 1972), 167-179. Stable URL: http sinks,stor.orgsici?sici=0021-8456% 28197221929 10%3A |23C 167% 3A ADAOPO%IE2.0,.COSB2-4 Journal of Accounting Research is currently published by The Instcute of Professional Accounting, Graduate Schoo! of Business, University of Chicago. Your use of the ISTOR archive indicates your acceptance of [STOR's Terms and Conditions of Use, available at tlp//ewvejstororglabouttenms.html. ISTOR's Terms and Conditions of Use provides, in pat, chat unless you fave obtained pcior permission, you may not dowaload an cnt isus of @ journal or multiple copies of articles, and you may use content inthe ISTOR archive only for your personal, non-commercial uss. Please contact the publisher cegarding any further use of this work. Publisher contact information may be obtained at bup:orww stor org/joummals/grad-uchicago.hen, Each copy of any part of a JSTOR transi printed page of such transtnission. ion must contain the same copyright notice that appears on the screen or ISTOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding ISTOR, please contact support @jstor.org- bup:thrwwjstor.orgy Sat hun 19.05:48:31 2004Research Reports A Discriminant Analysis of Predictors of Business Failure EDWARD B. DEAKINS ‘The failure of a business firm is an event which can produce substantial losses to creditors and stockholders, Therefore, @ model which predicts potential business failures as early a8 possible would serve to reduce such losses by providing ample warning to these interested parties. This was cufficient: motivation for Beaver (1967, 1968) and Allman (1968) to de- velop models for predicting failure based on the financial reparts of firms ‘The purpose of this paper is to propose an alionative model for predicting failure. Beaver used a dichotomous classifiestion test: to determine the error rates a potential creditor would exnetience if be classified firms on the basis of their financial ratios as failed or nonfailed, Beaver was able to accurately classify 78% of his sample of fins five years hefore failure, ‘Altman used discriminant analysis to rani firms on the basis of = ‘weighted combination of five ratios, His results were 95% effective in selecting future bantcrupts in the yesr prior to bankruptey. ‘The fiesas he ‘examined went bankrupt on an averege of seven and one-half months aiter the close of the last fiscal year for which reports were prepared. However, the predictive ability of the model declined rapidly as the number of years prior to failure incrensed. Tn the second throgh fifth years prior to failure, the discriminant model Jed to more misclassis- cations than did Beaver's dichotomous test using only @ eash flow /tatal debt ratio. ‘The correct classification rates from the two studies are eum- ‘marized in Table 1 Although Besver’s empirical results suggest that his method has greater predictive ability, the method used by Altman has more intuitive anpeal In this study, T will first reptieate the Beaver study, using the same ratios hhe used. Next I will search for the linear combination of the 14 ratios used Assistant Profowor, University of Texas, Austin. 187168 sOURNAL OF ACCOUNTING RESEANCH, SPRING 1972 TARLEL Classification Brror Rates for Predicting Future Rankrupteias eg afore Degrees ‘tena dain “ate sew tee ‘ante 1 1a 5 2 a 8 3 B 2 « x» 7 5 2 a Source: Beaver (1968), Altman (1968). by Beaver which best predicts potential failure in each of five years print to failure. Finally, I will devise a decision rule which will be validated over a erost-seetional sample of firms. The Dichotomous Classification Test ‘The dichotomons classification test, used by Beaver was based on two samples of firms, The first consisted of 79 firms which went bankrupt between 1954 and 1964. The second sample contained a nonfailed firm of approximately the same size and industry group for cach of the failed firms. Fourteen financial ratios were then ealevlated for each firm. ‘The ‘were selected on the basis of their appearance in the literature as tors of the ability of « firm to avofd failure. The overall set of firms ‘was then divided into two subsamples, each consisting of about half of the pairs of firms, ‘The firms in each subsample were then ranked by the ‘values of their ratios, That value of each ratio in one subsample which showed the smallest number of misclassifications was then used as the critioal value of the ratio for classifying the firms in the second subsam- pile. The procedure was then reversed by deriving © critical value of each ratio in the second subsample and using: that value to elassify the firms in ‘he first subsample, The number of errors resulting fram the use af each of the ratios in each of the five years before failure was calculated, and the better predictors were deemed to he those which showed the smallest classification error rate, ‘The same genetal procedute was followed in this study, but with one major difference. Thirty-two failed firms wete selected from a population ‘which experienced failure between 1964 and 1970. However, the term jailure was defimed here to include only those firms which experienced bankruptey, insolveney, ot were otherwise liquidated for the bonefit of creditors, Beaver included firms which defaulted on loan obligations or missed preferred dividend payments, But unless the nonbankrupt firms were matched by debt structttre as well as by size and industry, there could be a potential bias in eertain of the ratios. Fach of my failed firms ‘was matched with a nonfailed firm on the basis of industry classification, Yyeat of the financial information provided and asset sire,[PREDICTING BUSINESS FATIZRR 169 ‘The results of applying the dichotomous elassification test to this soce ond set of data are summarized in Table 2 along with the results obtained by Beaver. The data for this siudly wete not sufficient to determine if the differences in the error rates are significant. However, the Spesrman rank-order correlation coefficient can be used to indieate the order of the predictive power of the ratios in the two studies, The value of the coefi- sient (re) is indicated in Table 2. A review of the table indicates a rather high corvelation of relative predictive ability of the various ratios, The rauk-order eorrelation coeffi- cients are very high in four of the five years, Considering that differences ‘eould arise from the tse of independent samples and from the later time period of the second sample, the results wotld tend to confirm Beaver’s ‘observations. “However, the correlation coefficient in the thind year before failure is only .56, Although this is still a significant value, it is considerably lower than the coefficients in the other four years. & comparison was made of the ‘means of 13 financial statement items which were used in the calculation ‘of the financial ratios in the hope of giving some insight into why this coefficient is relatively less, These are shown in Table 8. Consider the aneane of total aseets for both sets of firms for the five years. Apparently, PARLE 2 Percentage Error for 14 Ratios on Dichotomous Classification Test Non-liguid amet group? ‘eae aw /total debt a ga) | a ae | a8 ca} | v6 a1) | 29 a5) neh income/totel assets 8 @) | 41 @9) | a0 ca) | 99 a) | 27 aay total debi/total sxsets 38 (28) | 99 27} | 22 C24) | 22 (2s) | a9 a9) Liquid asset to total asset group: ‘current asgeta/total amiees | 47 49) | 4B (a7) | AL G68) | 47 (48) | 36 8) ‘quick assers/total ansets 52 (a) | 52 (48) | 47 Gs) | 83 (2) | 34 GA) oriking aaplcal/total aasete | 84 al} | $0 45) | 8¢ Ga) | 28 «BA | 90 2s) achy/eotal assets 497-8} | 44 (8) | 47 GO) | 42 0} | 28 8) Liquid asset to eurzent debt group: cenrront ussets/eurrent liabilite ‘Guid aseeta/eurrent Kabilities ash /eurrent liabilitien Liquid asset turnover group: ‘current assets/sales ‘quick aesors/enlas ‘working coplcaleates ‘cash sales 40 5) | 36 8) | 25 6) | 27 <2) | a8 en) 4 a7} | 29 G4) | 41 40) | 28 G2) | 20H) 6 G8) | 45 (a8) | 96 a5) | 25 aa) | 96 22) ‘50 ct} | 56 49) | 82 G48) | oF Gon) | 38 Gay 153 (44) | 50 G2) | 85 Gs) | 60 7) | 52 0p 4 (ao) | 52 (48) | 45 (42) | a1 G9) | 28 Cy 48 (45) | 58 (43) | 52 Ge) | 42 2 | 47 GH) wo | me | om | ae Dats in parentheses are from Beaver (1068, p, 118)170 rDWARD 3, DaKEEN: TABLES Concparivon of Moana of £8 Financial Siatomont Items for Nonfaitel (NF) and Foiled @) Firms hea ‘Yn al ore 5 7 2 3 t Bales: NF 16989, 16305, 17610, aoe, F 16508, 1525. 1850. aes. eg. Dif. 1 amo, 740.8. [Nat income: NF 6, Ma, e rn 12s. ms. ual, equals the poptilation variance-covati- ‘ance matrix. To the extent thet these assumptions are not. teuo, the proba- Ditties will be incorrect, To examine the outeomes of the classifieation ‘esis, 1.260 classifications were obeerved and the assigned probabilities of group membership noted. The classifications were arrayed in the order of ‘the mersbership probability assigned, and the number of errors for each probability clase wes noted. The result of this test is summarized in Table AL. The expected error rates were derived by taking the mean of the TABLE At Claasifention Breors Separated into Aesiqnad Probabilities of Group Membership for 1,800 Selected Perm Classfcatins eet Macher racbatilty — Narber ofc x on 951.00 08, 7.2% 2.5% 90-949 38 na 75 ee 195 was 15.0 ToT 208 2a 250 0-609 2a 3010 36.0 [0-30 2 46.7 45.6PREDICTING BUSINESS PAILIEE — 179 slassification interval, and the observed error rates include all the misclas- sifications that ocourred where the assigned probability of gcoup metuber- ship was within the interval, “Phe results of this analysis tend to confirm the assumptions, although within some region of tolerance. Tt appears that the scotes of these firms tend to be more heavily concentrated in the tails of the density function ‘than is the case with ® normally distributed population. However, the ‘ertot rate for all firms within the probability class of 90 ta 1.00 would still be about 91.89%, which is not radically different from the interval tmoan of 95%. Therefore, tho difference botween the given probability and ‘the actual probability is probably small. We ean conclude that the diserim- inant olassification is sufficiently robust: to be used for distributions of financial data. ‘REFERENCES Avsunu, Howse, “Financial Ratios, Diserimizant Anulysis aad tho Prodietion of Corporate Bankruptcy.” Journal of Finance (September, 1068), np. 889-610 Asnencox, T. W. An Intraduetion to Multivariate Statitisa Analysis. New Yor obo Wiley, 1958 Bares, Went H. “Financial Ratioe as Predictors of Failure" Empirical Research ti Accounting: Selected Studies, 1907, supplement Lo Vol. 8, Journal of Accounting Research, pp. 71-102. —. “altemative Financial Ratio aa Prodietars af Failure,” The Accounting Review, XX Fonusry, 1988), 7-111 Brows, G. W. “Disctiminant Functions,” Annole of Mathematical Satirties, 18 (2947), sie, Pura, Wows G, ao Joaux J. Warowor. “A Proliction Model for Convertible Debeninres" Journal of Accounting Research, 9 Spring, 3971), 116-26. Pursmore, Pat, “An Bnpicieal Investigation of the Concept of Mateislty,” Bmpiricat Resoarch ‘x Accounting: Selected Studies, 1970, supplement '9 Vol. 8, Journal of Accounting Research, pp. 116-87. Bao, CB. Advanced Station! Methods in Biometric Retearch, New York: John Wiley, 1952, — "The Utilization of Multiple Measerements in Problems in Biological Clansii= cation," Journal of the Royal Statistical Sociaty, Series B (1948), pp. 160-88, “Touts, M, “Winaneial Ration ad 2 Means of Ferccusting Banlcuptey,” Maxagement International Rewiew,4 (1965), 15-21, ‘Tarsvona, Mavwten M. Discriminant Analycis, The Study of Group Diference. Chempaig, Iljnois: lastieute for Personality and Ability Testing, 197. wavs. Multivarato Anciwsic: Technique jaz Hducational and Peychological Resoareh. ‘New York: Joba Wiley, 171. ‘Winns, $.8,, Mathomatial Statistion. New York: Joba Wiley, 1952.