used by practitioners in the U.K. investment community, had accuracies of 96%,
70%, 61%, and 35% for the four years prior to failure.
The nearly perfect one-year-prior accuracy that T&T observe utilizing their model
contrasts sharply with the relatively small percentage of quoted and unquoted firms
that were assessed to have a going concern problem by their auditors. In fact, T&T
report that just 22% of the 46 quoted firms (and none of the 31 unquoted manufac-
turing bankrupt firms) had been qualified on-going concern grounds prior to failure.
(d) Implications. The drop-off in accuracy is quite noticeable as earlier year data
are applied. For investment purposes, however, one needs less of a lead time, versus
credit risk models, before failure in order to disinvest without losing a major amount
of his investment. It is fair to say, however, that as failure approaches, stock prices
tend to move downward in a rather continuous manner. Taffler and Houston (1980)
indicated that 12% of large quoted industrial firms had Z scores indicating high fail-
ure risk. This is a comparable figure to results we observed utilizing our own ZETA
model (Altman, Haldeman, and Narayanan, 1977) in the United States.
The authors also point out that about 15% to 20% of those firms which display a
profile similar to failed companies will actually fail. In addition, the British govern-
ment appeared to them to be keeping many ailing firms alive. Although this type of
paternalism is less common in the United States, examples like Lockheed and
Chrysler Corp. periodically crop up. Finally, T&T conclude that accountants are too
defensive when it comes to considering the value of conventional published historic
statements. When several measures of a firm, described from a set of accounts, are
considered together the value of the information derived is enhanced dramatically.
Essentially, T&T advocate a multivariate approach to financial analysis, and we cer-
tainly agree. It is unfortunate that they did not share with readers a more complete
description of their findings and the data used in their analysis. Their results are cer-
tainly provocative and appear to be of some practical use in England.
In his latest attempt to revise the company failure discriminant model (Taffler,
1982), a smaller sample of 23 failed companies (1968–1973) and 45 nonfailed enti-
ties displaying financially healthy profiles were examined first within a principal
component analysis framework. A large list of almost 150 potential variables was re-
duced to just five. These five are:
- Earnings before interest and taxes/total assets
- Total liabilities/net capital employed
- Quick assets/total assets
- Working capital/net worth
- Stock inventory turnover
The variables were discussed in terms of their discriminant standardized coeffi-
cients and other relative measures of contribution, but no function weights were pro-
vided. Taffler did utilize prior probability and cost-of-error estimates in his classifi-
cation procedures. He concludes that such an approach is best used in an operational
context as a means of identifying a short list of firms that might experience financial
distress (p. 15). Another conclusion is that the actual bankruptcy event is essentially
determined by the actions of the financial institutions and other creditors and cannot
strictly be predicted by using a model approach.
10 • 14 BUSINESS FAILURE CLASSIFICATION MODELS