10.20 FINLAND
(a) Suominen (1988). The author employs a multinomial logit model (MNL) to
classify firms into two groups: failing and nonfailing and to assess the relative im-
portance of each financial ratio variable. The second part of the study classifies failed
firms further into two groups: firms failed within one year of prediction and firms that
failed later. Both models employ the same set of three financial ratios indicative of
profitability, liquidity and leverage. The ratios are:
PROF(Quick flow – Direct taxes)/Total assets, where Quick
flow(Net turnover – Materials and supplies – Wages
and salaries – Rent and leases – Other expenses + Other
revenues)
LIQUQuick/Total assets, where Quick (Current assets –
Inventories/Current liabilities)
LEVELiabilities/Total assetsThe author favors the MNL technique, corrected for the constant term, because
concerns that the assumptions of equal covariance matrices and normal distribution
of the variables are not usually prevalent or tested when using discriminant analysis.
In addition, the coefficients from a MNL model are easily testable. Suominen’s sam-
ple consists of two sets of data. The first set covers the period 1964–1973 and con-
sists of 49 failed firms and 87 healthy firms, both from manufacturing industries. The
second set consists of data for a different set of failed and healthy firms covering the
period 1981–1982.
The PROF ratio was not found to be significant in the models for one and two
years prior to failure. In the three years prior model, only LEVE was significant. In
the four years prior model only LIQU was significant. The classification results on
the first sample and the second sample are summarized in Exhibit 10.18. It should be
noted that both results are for the sample space and not for holdouts. The results of
the one-year model are comparable to those obtained using discriminant analysis
using the same variables. The Type I errors are reported to be fewer in the discrimi-
nant model, however.
The purpose behind the second part of the study is not entirely clear. Here the ob-
jective is to predict correctly the firms that failed within one year of the prediction as
10 • 42 BUSINESS FAILURE CLASSIFICATION MODELSData From 1964–1973 Data From 1981–1982Years Type I Type II Type I Type II
Prior Accuracy % Accuracy % Accuracy % Accuracy %
1 67–71 85–86 65–74 61–65
2 53–57 84 61 70
3 31–33 87–89 65 70
4 26 93–95
Exhibit 10.18. Classification Accuracy.