The study on bankruptcy come to the 1930's beginning with the use of ratio analysis to predict future bankruptcy up to the mid-1960's based on single factor/ratio analysis usually used for comparison purpose. The most widely recognized univariate study (single factor/ratio study) is that of Beaver (1966).
Beaver (1966) demonstrated for first time ability of ratios to predict individual firm failures and found that the cash flow (CF) to debt ratio performed best as a signal of impending financial failure. He used a dichotomous classification test to predict the failure status of a firm to determine the wrong judgment a potential creditor would experience if he placed firms on the basis of their financial ratios as failed or non-failed. For this purpose he used six different classes of ratios, named as cash-flow ratios, liquid-asset to current debt ratio, net income ratio, turnover ratio, debt to total assets ratios and liquid assets to total assets ratios. In all these groups he calculated 30 ratios based on two samples of firms. The first sample consisted of 79 firms which went bankrupt between 1954 and 1964. The second sample contained a non-failed firm of approximately the same size and industry group for each of the failed firms. He included firms which defaulted on loan obligations or missed preferred dividend payments. Three criteria were used to select the 30 ratios which were popularity-frequent appearance in the literature, ratios performed well in one of the previous studies and ratio be defined in terms of a "cash-flow" concept. To see the predictive ability of the model, Beaver arrayed the data (i.e., each ratio is arranged in ascending order and found an optimal cutoff point) that classify the failed and non-failed status of the firms. According to Beaver ability to predict failure status is strongest in the cash-flow to total-debt ratio. He was able to accurately classify 78% of his sample of firms five years before failure.
Neter (1966) made his remarks on the Professor Beaver (1966) work, and he raised many question related to firms selection criteria, selectivity criteria of variables, and dichotomous classification test application as compared to other naive models. He raised questions like; were the factors that were used for matching the most effective?, were the ratios used for analysis are more effective, and what is naïve model as compared to Beaver (1966) model.
In 1968, Altman published the first multivariate study, which remains very popular in the literature till today. After this, a variety of studies in bankruptcy prediction studys come across, those all studys have different models and different factors both in numbers and variety. For example, Altman's (1968) model is a five-factor multivariate discriminant analysis model while Jo, Han and Lee's (1997) model is a 57-factor. Discriminant analysis was a very popular method for model development in the early stages of bankruptcy prediction but advancements in technology come up with other methods including logit analysis, probit analysis, and neural networks are more prominent but most widely used method remains discriminant analysis (More than 67 studies).
Since 1968, the major methods that have been used for model development are multivariate discriminant analysis (MDA), logit analysis, probit analysis, and neural networks. But as discussed earlier multivariate discriminant analysis (MDA) is widely used in more studies than other methods. Multivariate discriminant analysis (MDA) is used for predicting bankruptcy of the firms by Altman (1968, 1973, 1978, 1984, 2000, 2002), Edmister (1970), Deakin (1972), Edmister (1972), Gru (1973), Blum (1974), Sinkey Jr. (1975), Joy and Tollefson (1975), Huberty (1975), Tisshaw (1976), Moyer (1977), Altman, Baidya and Ribeiro Dias (1979) , Ohlson (1980), Collins (1980), Taffler (1982), Chen and Shimerda (1981), Taffler (1984), Ketz (1978), Mason and Harris (1978), Norton and Smith (1979), Sharma and Mahajan (1980), Collins (1980), Altman and Spivack (1983), El Hennway, Morris (1983), Mensah (1983), Izan (1984), Ambrose and Seward (1984), Grammatikos and Gloubos (1984), Takahashi, Kurokawa and Watase (1984), Rose and Kolari (1985), Katz, Lilien, and Nelson (1985), Gombola, Haskins, Ketz and Williams (1987), Goudie (1987), Moses and Liao (1987), LAU (1987), Goudie and Meeks (1991), Papoulias and Theodossiou (1992), McGurr and DeVaney (1998), Hsieh and Wang (2001), Massman et al. (2001), Neophytou, Charitou and Charalambous (2001), Grice and Ingram (2001), Gu (2002), Jones and Hensher (2004), Agarwal and Taffler (2007), Bandyopadhyay (2006, 2007), Jayadev (2006), Gerantonis, Vergos and Christopoulos (2009), Miller (2009), Samarakoon and Hasan (2009), Wang and Campbell (2010), and others in their studies.
Altman (1968) used discriminant analysis to rank firms on the basis of a weighted combination of five ratios. The basic concept behind this study is that why we only used ratios for comparison purpose, these can be used for decision making in more sophisticated statically way than just compare one firm to another. Altman want to address many question in his study, major questions are, how we can used ratios for decision making, if we used the ratios, which ratio had more importance than the other which help in more bankruptcy detection potential, what weight should be attached to those ratios, and how should weights were objectivity established to fulfill required objectives. He used multi- discriminant analysis for this purpose. According to author, multi- discriminant analysis fulfills the requirement although it is not as famous as regression analysis. In MDA author have to select two groups e.g. bankrupt or non-bankrupt. A sample of sixty six firms was selected at the start with thirty three firms in each two groups from manufacturing industry. The basic problem raised during sample selection was size effect, but according to the author financial ratios by their nature have effect of deflating statistics by size so size effect is eliminated. Altman selected 22 potential ratios for his model development, which were groups into five different categories like liquidity, profitability, leverage, solvency and activity ratios. These ratios were selected on three main criteria's, popularity in literature, potential relevancy to the study, and few new ratios appearing in literature. Out of twenty two ratios, five ratios were finally selected after utilizing four different criteria's, those criteria's were, statistical significations of ratios, inter-correlation between ratios, predictive accuracy of different profiles and judgment of analyst itself.
Altman was resulted final discriminant function as follows:
Z = .012X1 + .014X2 + .033X3 + .006X4 + .999X5
Where
X1 = Working capital/Total assets
X2 = Retained Earnings/Total assets
X3 = Earnings before interest and taxes/Total assets
X4= Market value equity/Book value of total debt
X5 = Sales/Total assets
Z = Overall Index
Final results for study were 94% correct prediction at initial level while overall 95% effective in selecting future bankrupts in the year prior to bankruptcy. The firms examined went bankrupt on an average of seven and one-half months after the close of the last fiscal year for which reports were prepared. The limitation of study was that the firms examined were all publicly held manufacturing corporations. According to the author, accuracy of model falls off consistency with exception of the fourth and fifth years. Altman concluded his study that Z score of model greater than 2.99 clearly fall into bankrupt sector, 2.99 to 1.81 are in zone of ignorance, and less than 1.81 came in bankruptcy stage.
Deakin (1972) used a dichotomous classification test and discriminant analysis to rework the Beaver (1966) and Altman (1968) results in his study. For dichotomous classification test verification, thirty-two failed firms were selected from a population which experienced failure between 1964 and 1970 but selection criteria include only those firms which experienced bankruptcy, insolvency, or were otherwise liquidated for the benefit of creditors. Although Beaver's claimed that his observed results proposed that his method has greater predictive ability, but this study results signified both models, and he conclude that discriminant analysis (Altman 1968) is sufficiently vigorous to see predictive ability of the data more accurately than the other method.
Altman (1973) used linear discriminant analysis to identify and quantify those financial measures which are effective indicators of bankruptcies to predict bankruptcy of railroad of America. A group of twenty-one railroads that went bankrupt between the years 1939-1970 was selected for this purpose as compared to same size of non-bankrupt group. Three different groups of financial indicators, liquidity measures profitability and efficiency measures and solvency and leverage measures are taken for further analysis of data. Fourteen ratios were selected from these three groups for analysis; data is taken from moody published reports on railroads. In first step of analysis, author calculated averages for those 14 ratios for the bankrupt group sample of 21 firms based on data drawn from one statement and from two statements earlier to bankrupt. The bankrupt group's ratios showed significantly worse measures than the industry averages for both one and two statements previous to failure. These results are on the basis of F-ratios significant at 0.01 levels. From all of the further analysis, author came up with this equation;
Z = 0.2003(X14) - 0.2070(X7) + 0.0059(X11) - 0.0647(X1o) + 0.1040(X8) + 0.0885(X6) + 0.0688(X3).
X14 = cash flow/fixed charges
X7 = transportation expenses/operating revenue
X11 = earned surplus/total assets
X10 = three-year compound growth rate in operating revenues
X8 = earnings after taxes/operating revenues
X6 = operating expenses/operating revenue
X3 = earnings before interest and taxes/total assets
The group means of the two-group railroad sample are (1) Bankrupt Railroads = - 3.640 (2) Industrial Average's = + 0.299. In the calculation of Z-scores for each observation, it is important to note that the first variable (X14) is expressed as number of times, e.g. 2.3, while the other six variables are all expressed in "absolute" percent form. Empirical results showed that railroad sample is extremely accurate with 97.7 percent correct classification at one and two years prior to bankruptcy. The cutoff point for the model is - 1.465, the less value of Z-Score than this value showed bankruptcy of that firm and vice versa.
Moyer (1977) re-examined the Altman (1968) study after criticized by Joy and Tollefson (1975) in their study "On the Financial Applications of Discriminant Analysis". They criticized the Altman (1968) study into different aspect mainly the absence of ex-ante validation, the criteria used in selecting variables for inclusion in the model, and the lack of appropriate comparisons with naive alternative models. According to this study as Altman (1968) model has much importance in literature and text book of financial risk management etc, so there is a need of re-examine this model after this criticism. Moyer (1977) first checked the predictive power of the Altman (1968) model, applied to a new set of data from firms during the 1965-1975time period and then Altman model parameters are re-estimated based upon the new sample and at the end the explanatory power of the re-estimated model is compared with an alternative model based upon the univariate studies of Lev and Beaver. Study used data of 27 bankrupt and 27 non-bankrupt firms during the 1965-1975 time interval for its analysis of re-examined. When he applied data to original Altman model, he found overall success rate of 75% of this model instead of 95% of Altman results. In re-estimating the entire model and its variable, this study used both Direct and WILKS approaches. After application and analysis, study found that overall predictive power of the model is increased by 88.01% when used direct methods, and 90.48% when used WILKS approach. Moyer (1977) found two equations, one from direct methods, and other from WILKS,
Altman by DIRECT: Z = .314 - .0059X1 - .0053X2 - .OO33X3 - .OOOO5X4 - .00023X5
Altman by WILKS: Z = .814 - .019X1 - .017X2 - .0097X3
According to Moyer (1977), Joy and Tollefson (1975) were corrected that Altman had used an improper test of the significance of his variables, but when we compared this against another model, it has more predictive ability.
After Moyer re-examination study, Altman (1978) replied to that re-examination, and done further examination of that re-examination of Moyer study. Altman replied to Moyer, that it is correct that "Proof is in eating", but to test the model with its all assumption is also necessary for its validity. Altman also accept the fact that any model will have useful predictive content if the populations are relatively stationary over time.
Altman (1978) in its reply to Joy and Tollefson's (1975) commented in this study, ex post classification provided a crude test of the stationary of the model and its individual component measures and parameters and ex post classification provides an indication of the confidence one can have with respect to the observed group overlap among the variable distributions in the groups being investigated. So this is in contrast to measures of variable significance which are based upon statistics that apportion the distance between group means.
According to Altman (1978), Moyer (1977) is unnervingly inaccurate in presenting his results of a sample of 48 firms over the period 1965-1975. This relatively small sample size (48 with 23 bankruptcies), as compared to the population of bankrupt firms in the sample period; hence one wonders how his sample was determined. When Altman re-examined the Moyer's study, he found that, out of total 27 bankrupts firms 26 firms are located for information. Several of the firms are in-appropriate for comparison purposes, including three retailers and two railroads. In all, 10 of the firms are non-manufacturers, and another was not a bankrupt. Of the remaining 15, 12 had Z-scores lower than the cutoff score. So Altman observed a 20% error rate. The error rate is somewhat higher than the previous studies but certainly lower than Moyer's 39% assessment. So one can easily say that re-examination of Moyer (1977) has many faults in choosing the appropriate sample. Altman also replied to Moyer about comments of using best model in each year for prediction, this is debatable but Altman (1978) caution that the use of separate models is confusing, oftentimes misleading, and cannot establish a precise time-to-bankruptcy likelihood as Moyer (1977) claims. In concluding remarks, Altman have every confidence that the Z-score model is still upwards of 80% accurate for business manufacturing failures that have occurred since 1965.
Altman et al. (1977) constructed a second-generation model with several enhancements to the original Z-score approach. The new model, which was called ZETA, was effective in classifying bankrupt companies up to five years prior to failure on a sample of corporations consisting of manufacturers and retailers. The ZETA model tests included non-linear (e.g., quadratic) as well as linear discriminate models. The non-linear model was more accurate in the original test sample results but less accurate and reliable in holdout or out-of-sample forecasting. Subsequently, in Altman et al. (1995) modified his Z-score model to emerging market corporations, especially Mexican firms that had issued Eurobonds denominated in US dollars. In this enhanced Z-score model, he dropped sales/total assets and used book value of equity for the fourth and final variable to make it more suitable for the private firms.
Altman, Baidya and Dias (1979) examined in this study about the recent business failures in Brazil for classifying and predicting serious financial problems of companies. They utilized a bankruptcy classification model developed by Altman (1968) in order to classify Brazilian firms during the period, 1973 to 1976. A sample of 23 serious-problem firms is compared with same number of slightly larger control sample of healthy firms. Their four-variable model successfully classified 88 percent of the firms one year prior to serious problems and as much as 78 percent three years prior. The importance of models that highlight financial problems early enough to suggest remedial changes is apparent in Brazil.
They discussed empirical results in terms of two separate, but quite similar, models. The first model, referred to as Z1, includes variables X2 . . . X5 (four measures) but Model Z1 does not include X1 while the second model, referred to as Z2, does not include X2, in both cases, the critical cutoff score is zero, both final models are as given;
Z1= -1.44 + 4.03 X2 + 2.25 X3 + 0.14 X4 + 0.42 X5
Z2 = -1.84 - 0.51 XI + 6.23 X3 + 0.71 X4 + 0.56 X5.
The number of observations was reduced from twenty-three in year one, to nineteen (year two) and eighteen (year three).
Altman and Spivack (1983) discussed in their study compared two of the more prominent measures of corporate financial viability available to the investment community-the Zeta model of bankruptcy classification and the Value Line Relative Financial Strength System. Both the Zeta model of bankruptcy classification and the Value Line Relative Financial Strength System classify public corporations according to characteristics of financial health. Zeta model used financial variables to discriminate between bankrupt and non-bankrupt firms, whereas Value Line relates similar types of variables to the observed yields of outstanding debt securities. Zeta is essentially an objective system; Value Line's ratings are a function of both mathematical relations and human judgment. Both models employed measures of profitability, leverage, size, variability and market value. Zeta, however, measured earnings variability, whereas Value Line evaluates stock price stability. Value Line's size measure involves earnings; Zeta model used total assets. Despite these differences, a comparison of the two systems revealed that Value Line's scores exhibit a high correlation with Zeta scores, which have been shown to discriminate well between bankrupt and non-bankrupt firms. Both systems scores also correlated well with published bond ratings.
Altman (2002) discussed and reviewed two of the more prominent credit scoring techniques, Z-Score and KMV's EDF models in this study. This study primarily discussed a model developed by the author over 30 years ago Z-Score model, and its relevance to these recent developments and an alternative widely used credit risk model, known as the KMV approach and they also compare both KMV and Z-Score in the now well-known Enron (2001) bankruptcy disaster. He discussed all the development procedure of Z-Score through MDA analysis as discussed in Altman (1968) study. They come up with three equations of Z-Score which are;
Z = 1.2 X1 + 1.4 X2 + 3.3 X3 + 0.6 X4 + 1.0 X5
(X1 = working capital/total assets, X2 = retained earnings/total assets, X3 = earnings before interest and taxes/total assets, X4 = market value equity/book value of total liabilities, X5 = sales/total assets, and Z = overall Index or Score where 1.81 cutoff point)
For Private Firms Z' = 0.717(X1) + 0.847(X2) + 3.107(X3) + 0.420(X4) + 0.998(X5)
Non-Manufacturers and Emerging Markets Z" = 6.56 (X1) + 3.26 (X2) + 6.72 (X3) + 1.05 (X4)
They applied these two models on the bonds and showed their predictive ability over the different kinds and categories of bonds is correct over different years. In concluding remarks they said that Enron case and many others although tools like Z-Score and EDF were available, losses were still incurred by even the most sophisticated investors and financial institutions so having the models is simply not enough, the actual thing that needed is a "credit-culture" within these financial institutions, whereby credit risk tools are "listened-to" and evaluated in good times as well as in difficult situations.
Edmister (1972) worked on small business failure prediction by samples of 562 and nineteen ratios are calculated from the borrowers' balance sheets and income statements. The ratios were divided by matching industry-size composite ratios. Stepwise multiple discriminant analysis with restrictions upon inter-correlation was employed to estimate a function and found 99% accuracy of model prediction. Author also concluded that the predictive power of ratio analysis is cumulative mean single variable predicted failure nearly as well as a small group of variables and some variables which were not significant predictors alone added discriminatory ability to a function containing selected other variables.
Blum (1972) developed a model to predict failure for firms in general. He used discriminant analysis to test the hypothesis which can distinguish between failing and non-failing firms. The analysis was applied to a paired sample of 115 failed and 115 non-failed firms which failed during the years 1954-68. The model distinguished failing from non-failing firms with an accuracy of approximately 94 percent, when failure occurred within one year from the date of prediction, 80 percent for failure two years into the future, and 70 percent for failure three, four, and five years distant.
Sinkey (1975) described a multiple discriminant analysis of the characteristics of problem banks. He analyzed 110 problem banks identified in 1972 and 1973 for this purpose. He employed financial ratios derived from balance sheets and income statements for the years 1969-1972. The financial ratios are designed to measure the bank's performance in such areas as liquidity, efficiency, and capital adequacy. He used ten-variable set for his analysis. Chi-square results indicated that for each of the years 1969-1971 the mean vector of the problem bank is expected to be more like the mean vector of the non-problem bank, and vice versa, than are the vectors of 99.55% of individual group members.
Altman, Baidya and Ribeiro Dias (1979) examined in this study, the recent business failure experience in Brazil and developed a model for classifying and predicting serious financial problems of companies in order to classify Brazilian firms during the period, 1973 to 1976 with a sample of 23 serious-problem firms with compared of healthy firms. A four-variable model successfully classified 88 percent of the firms one year prior to serious problems and as much as 78 percent three years prior.
Collins (1980) sought and compared two models, one of Altman (1968) and Meyer and Pifer (1970). In order to compare these two models, he used a large sample of 162 bankrupt and same number 162 non-bankrupts of chartered credit unions from 1956 to 1976. Collin used 9 ratios X1 = cash + securities divided by total assets as a measure of liquidity; X2 = net income divided by total assets as a measure of profitability; X3 = delinquent loans divided by total loans as a measure of the likelihood of future asset losses; X4 = new loans divided by total loans as a measure of activity; X5 = reserves divided by risk assets as a measure of the ability to withstand asset losses; X6 = dividend rate; X7 = number of potential members; X8 = number of actual members; and X9 = TL assets (measured in thousands). The results showed Altman and the Meyer and Pifer methods of bankruptcy prediction provided good predictive results, Altman method performs just as well or better than the more theoretically appealing Meyer and Pifer method.
Sharma and Mahajan (1980) analyzed failures and propose a failure process model. The model proposed that failures can be predicted by analysis of either the causes of failure or the performance indicators. They developed failure model by taking sample of 46 firms from retail industry. Multi discriminant analysis was conducted to find the "best" linear discriminant functions for the five years prior to failure. According to author current ratio and return on assets are two important indicators of business failure.
Ohlson (1980) in this study presented some empirical results of a study predicting corporate failure as evidenced by the event of bankruptcy. The data set used in this study is from the seventies (1970-76). I know of only three corporate failure research studies which have examined data from this period. Study relied on observations from 105 bankrupt firms and 2,058 nonbankrupt firms from 10-K financial statements. He used econometric methodology of conditional logit analysis to avoid some fairly well known problems associated with Multivariate Discriminant Analysis. According to author there are certain statistical requirements imposed on the distributional properties of the predictors' i.e, the variance-covariance matrices of the predictors should be the same for both groups (failed and nonfailed firms); moreover, a requirement of normally distributed predictors certainly mitigates against the use of dummy independent variables. The results showed that size, financial structure (Total Liabilities to Total Assets), performance and current liquidity were important determinants of bankruptcy. In the logit analysis, average data is normally used and it is considered as a single period model. Hence, for each non-distressed and distressed company, there is only one company-year observation.
Taffler (1982) described in his study discriminant model for the identification of British companies failure and discussed the results from its application since it was developed and its degree of inter-temporal validity. He developed failure model by taking sample of 23 failed firms and 45 non-failed firms from retail industry. He used five ratios: earnings before interest and tax/opening total assets (EBIT/TA-1), total liabilities/net capital employed (TL/NCE), quick assets/total assets (QA/TA), working capital/net worth (WC/NW) and stockturn. Their model achieved prediction accuracy rates of 96%, 70%, 61%, and 35% for 1 year to 4 years prior to failure.
Goudie (1987) have developed a distinct framework designed to identify those members of the corporate sector that will experience severe financial difficulty, potentially leading, ceteris paribus, to failure. Our preliminary application of the model to a restricted sample revealed a promising discriminative ability, with average type I and II errors of 17.9% and 5.7% respectively for up to five years forewarning. The procedure led to estimated type I and type II errors of 20.8 % and 10.7 % respectively, giving an overall correct classification of 84.7%.
Grammatikos and Gloubos (1990) worked on Greek firms for prediction of bankruptcy model. They used Multiple Discriminant Analysis and the Linear Probability Model in predicting financial crisis even three years prior to the eventual failure. For the MDA the overall successful classification rates are 91./%, 78% and 70% for the first, second and third years before bankruptcy. The corresponding classification rates for the LPM are 91.4%, 76% and 78%.
Papoulias and Theodossiou (1992) provided an analysis of recent business failures in Greece and presents models for detecting financially distressed firms. Logit, probit, linear probability model, and linear discriminant analysis are used for developing these models. They used 33 failed firms and 68 non-failed firms for this analysis. They included variables Current assets/current liabilities (CA/CL), Working capital/total assets (WC/TA ), (Current assets-Inventories)/current liabilities (QA/CL), Net income/total assets (NI/TA), Gross profit/total assets (GP/TA), Long-term debt/total assets (LTD/TA), Total debt/total assets (TD/TA). Results predicted results of goodness as LPM 85.2, BDA 85.2, Logit 85.2, Probit 80 85.2.
Fadhil, Ali and Harris (1994) described research directed towards the development of an operational system for identifying construction companies in danger of failure over 1978 to 1986 yearly data with 11 failed and 20 non-failed companies. They applied MDA method for analysis of data, by using seven variables. The models correctly 100% predicted failed firms while 67% non-failed firms.
Dimitras et al. (1996) extensively reviewed business failure prediction methods used by researchers and analysts documented in bankruptcy prediction literature. A total of 158 articles were reviewed and classified according to country, industrial sector, time period, financial ratios and models or methods used. Their review revealed that MDA was the predominant method used for business failure prediction and the most significant classifying ratios were in the solvency category. Like many previous studies, this study of restaurant bankruptcy classification also uses MDA because it is the most widely used bankruptcy prediction method and has been proven to be quite effective in making predictions.
Ismail (1997) proposed MDA model for predicting bankruptcy of firms in their countries like from Malaysia for Property Development Firms. According to author, "it is expected that the model proposed should help them to identify some of the model significant financial ratios shaping the financial performance of the property development firms in which they are involved".
Mossman et al. (1998) they compared four types of bankruptcy prediction models based on financial statement ratios, cash flows, stock returns, and return standard deviations are compared. Based on a sample of bankruptcies from 1980 to 1991, results indicate that no existing model of bankruptcy adequately captures the data. During the last fiscal year preceding bankruptcy, none of the individual models may be excluded without a loss in explanatory power. If considered in isolation, the cash flow model discriminates most consistently two to three years before bankruptcy. By comparison, the ratio model is the best single model during the year immediately preceding bankruptcy. For first year, the ratio model best classifies firms; this showed that 84.9% (83.9%) of all size-primary (industry-primary) firms are concordant with model allocation. The cash flow model follows closely, accurately classifying 84.0% (82.6%) of firms in sample when size (industry) matching is used. Return and variation models display poorer ability to discriminate between bankrupt and non-bankrupt firms during the last two years.
Shirata (1998) presented some empirical results of a study regarding financial ratios as predictors of Japanese corporate failure, evidenced by bankruptcy. The study proved that the model can predict bankruptcy with more than 86.14% accuracy regardless of industry and size. He used basic 61 ratios at start, and a data set for this study is 686 bankrupt firms and 300 non-bankrupt firms. He developed specific equation for bankrupt model. (Z = 0.014X2 - 0.058X24 - 0.062X36 - 0.003X10 + 0.7416)
In another study Grice and Ingram (2001) evaluated the generalizability of Altman's (1968) Z-score model using a proportionate sample of distressed and non-distressed companies from time periods, industries, and financial conditions other than those used by Altman to developed his model. The findings indicated that the accuracy of Altman's model declined when applied to 1985-1987 estimation sample of a 1988-1991. Altman reported an 83.5% overall accuracy for his model using a sample from 1958 to 1961. The overall accuracy for the 1988-1991 sample used in this study was 57.8%. Altman's model was sensitive to industry classifications in the sample used in this study. The overall accuracy of the model was significantly higher for manufacturing firms (69.1%) than for the entire sample (57.8%) that included non-manufacturing firms. Altman's model was not sensitive to type of financial distress. The overall accuracy of the model for bankrupt companies (56.1%) in the 1988-1991 sample was not significantly different from that of the entire sample (57.8%) that included other financial distress situations.
Another study estimated a multiple discriminant model for analyzing US restaurant firm bankruptcy. The estimated model classifies the in sample firms accurately with a 92-percent accuracy rate 1 year prior to bankruptcy, similar to those achieved in previous studies for other industries. He came out Z-score equation of Z =-1.215X+ 2.498X2 - 0.114 - 1.664X - 0.256 = -1.0738, where X total liabilities to total assets. The estimated model suggests that in the US restaurant industry, debt-burdened firms with poor EBIT are more likely to be candidates of bankruptcy. Zheng (2002)
Bandyopadhyay (2006) constructed bankruptcy model with several enhancements to the original Z-score approach (Altman 1968) for predicting corporate bond default in India. The new Z-score model developed in this study depicted not only a high classification power on the estimated sample, but also exhibited a high predictive power in terms of its ability to detect bad firms in the holdout sample. The model clearly outperforms the other two contesting models comprising of Altman's original and emerging market set of ratios respectively in the Indian context. He worked out and come up with three models; Model 1: re-worked Altman (1968), Model 2: re-worked emerging market (1995), Model 3: new Z-score model. The predictive ability of three models are 95.3% to 82.5% for first year to second year, for re-worked model the results are 95.3% to 85%, and for new Z-score model 96.3% to 87% from first to second model.
Another study by Agarwal and Taffler (2007) described a widely-used UK-based z-score model including publication of its ratio coefficients for the first time and explores its track record over the twenty five year period since it was developed. The study demonstrated that the z-score model depicted which was developed in 1977 has true failure prediction ability. The model is shown to have clear predictive ability over this extended time period and dominates more naïve prediction approaches.
They proposed some suggestions for development of Z-Score model which mainly are; The model developed should apply to a wide variety of firms, Z-score model component ratios need to reflect the current key dimensions of firm financial profiles, and ratio set and coefficient estimates need to be jointly determined, coefficients should be recalculated with new ratios.
Chung (2008) study employed financial ratios for differentiating between failed and non-failed financial companies (non-bank) in New Zealand. These financial variables were derived from the financial statements of both failed and non-failed companies. Methodologies adopted included univariate tests, MDA (stepwise regression) and ANN (back propagation algorithm). The univariate tests indicated that failed company's financial ratios differ significantly from non-failed companies. Failed companies were less profitable and less liquid. They also had higher leverage ratios and lower quality assets. The results of the optimal MDA model indicate that the models are more accurate with data one year prior to failure.
Gerantonis, Vergos and Christopoulos (2009) examined whether Z-score model, developed by Altman (1993) can predict bankruptcies or not. They found evidence that this model is useful in identifying financially troubled companies that may fail up to 2 years before the bankruptcy. Overall model success rate is not statistical significant, but when it comes to failure cases it can predict 54% of them one year before failure. The predictive ability of Altman model is in line with findings by other researchers in Greece and in the United States.
In a study, Gulsun and Umit (2010) have developed and tested a statistical early warning model to identify companies experiencing deteriorating financial health by examining 45 insurance companies acting in non-life elementary branches of insurance during the period between 1992 and 2006. They developed the model using data regarding 45 dependent and 17 independent variables and logit model Multi Discriminate analysis and Regression analysis. This study compared the ability of logit, discriminant and regression analyses to predict insurance company underperformance. When comparing the predictive ability of all three models, the logit model showed slightly better prediction ability than the other models. The three models used information from 2003 - 2006 to predict the performance of insurance companies in 2007.
Wang (2010) study used data from Chinese publicly listed companies for the period of September 2000-September 2008 to test the accuracy of Altman's Z-score model in predicting failure of Chinese companies. Prediction accuracy was tested for three Z-score variations: Altman's original model, a re-estimated model for which the coefficients in Altman's model were recalculated, and a revised model which used different variables. All three models were found to have significant predictive ability. The re-estimated model has higher prediction accuracy for predicting non-failed firms, but Altman's model has higher prediction accuracy for predicting failed firms. The revised Z-score model has a higher prediction accuracy compared with both the re-estimated model and Altman's original model. This study indicates that the Z-score model is a helpful tool in predicting failure of a publicly listed firm in China. The drived the new Z-Score model is Z=0.8059X1-0.2898X2 +0.0440X3 +0.1971X4 +6.3327X5, Firms with Z-scores less than 2.2373 are predicted to be delisted and Z-scores greater than 2.2373 are predicted to be non-delisted. The model correctly predicted 90% of firms (54 out of 60), with both type I and type II error at 10%. This is higher than 76.67% overall accuracy for the estimation sample using Altman's (1968) model. However, the Type I error is lower using Altman's (1968) model (3.33%) compared with their model. the revised Z-score model. Revised model Z=0.2086X4 +4.3465X5 +4.9601X6. The cut score third model is 1.5408. The revised model correctly classified 95% of the firms in the estimation sample. It misclassified 3 out of the 30 delisted firms and correctly classified all the non-delisted firms. The estimation sample overall accuracy rates of the revised model are 95% and 91.67% respectively for one year and two years prior to delisting. Boritz et al. (2007) also re-estimated the model using Canadian company data and obtained the following: Z=2.149X1-0.624X2+1.354X3-0.018X4+0.463X5
Lifschutz and Jacobi (2010) carry out investigation of whether Altman Model (1968) is good enough to predict financial failure of publicly traded companies in Israel between 2000 and 2007. The findings of the study indicated that the preferable model for predicting financial failure of Israeli companies is the Ingbar (1994) version of the Altman Model. Their results showed that model is able to predict bankruptcy of companies with a 95% accuracy rate one year prior to bankruptcy and with an 85% accuracy rate two years prior to bankruptcy than Altman(1968) model that predict 75% in first year and 65% in second year.
Ingbar, Y. (1994). Analysis of financial statements. Israel Institute of Productivity. (Chapter 13). [in Hebrew]
