3.Logit Analysis (LOGIT)
4.Multiple Discriminant Analysis (MDA)The LPM model is a multiple linear regression model where the dependent variable
is a 0–1 variable which is regressed against a set of independent variables. One prob-
lem with this approach is that the error terms’ distribution is not normal. Also when
the predicted value lies outside the 0–1 range, it is difficult to interpret the result. This
difficulty is overcome by applying suitable transformations that would restrict the
probability predictions to the 0–1 interval. This is done in the PROBIT model where
P is the conditional probability of failure expressed in terms of a cumulative standard
normal distribution function. As to be expected, the introduction of the standard nor-
mal distribution involved nonlinear estimation. The LOGIT model uses a computa-
tionally simpler function based on the cumulative logistic probability function. In
multiple discriminant analysis, the function is linear or quadratic in the variables.
The sample consisted of 30 Greek industrial firms that went bankrupt during the
period 1977–1981. Each failed firm was paired with a healthy firm of similar size in
the same year and from the same industry. Data was gathered for one year prior to
bankruptcy and was obtained from various issues of the Government Gazette. Sev-
enteen accounting ratios were used in the analysis and the final models with all four
techniques had the same variables. The group statistics for these ratios along with the
T-statistics are presented in Exhibit 10.10.
The model results on the development sample are as reproduced in Exhibit 10.11.
It was found that the MDA and LPM have the greater accuracy overall and also in
the Type I and Type II categories. The authors note that the MDA model’s coefficients
for two of the variables had counterintuitive signs but go on to suggest that because
of the interdependencies inherent in a multivariate model, this may be acceptable.
10.12 GREECE 10 • 29Group Mean Group Mean
Variable Bankrupt Nonbankrupt T-valueCurrent assets/current liabilities 0.932 1.579 –3.95
Net working capital/total assets –0.092 0.196 –5.20
Total debt/total assets 0.813 0.595 5.69
Gross income/total assets 0.077 0.253 –4.51
Gross income/current liabilities 0.106 0.607 6.16
Exhibit 10.10. Group Statistics.
A. One year prior to
bankruptcy Overall Bankrupt Nonbankrupt
MDA 91.7% 96.7% 86.7%
LPM 91.7 93.3 90.0
PROBIT 85.0 83.3 86.7
LOGIT 86.7 83.3 90.0
Exhibit 10.11. Correct Classifications on the Original Sample.