forced to use the absolute number of failures. The authors also note the other short-
comings: limited size of the data sample, conceptual problems with measuring ex-
pected devaluation rates, and the distortions in measuring the time of failure by lags
in court processing time. The authors conclude, based on the results of the regres-
sions, that of all the factors considered, interest rates and credit stocks are the most
important factors in explaining business failures.
The second question examined by the authors is the issue of whether all industries
wereuniformlyaffected by the Argentine reforms. The authors’ hypothesis is that the
high protection industries suffer considerably higher failure levels than the low pro-
tection industries when the protection is reduced. Each subsample for the study con-
sisted of 12 industries with data for 20 quarters. To account for interindustry differ-
ence in the number of firms, the authors included the logarithm of the number of
establishments in the industry as an explanatory variable. The authors report statisti-
cally significant evidence to support their hypothesis that high protection leads to
higher failures when protection is removed.
In order to test their third question, that is, what are the firm-level variables that
predict failure, the authors favor the use of a probit regression instead of discriminant
analysis for two stated reasons: that assumptions necessary for statistical inference
are typically not satisfied and that the individual influences of the predictors cannot
be isolated. The criticism of discriminant analysis by the authors is not compelling
because the authors appear to tolerate even more serious limitations caused by the
smallness of the sample. Also the standardized discriminant function does show the
relative importance of the variables.
The models were estimated for the predevaluation period and postdevaluation pe-
riod. The final model contained ratios with total assets as the best normalizing vari-
able (as opposed to total debt or net worth). The resulting model included the fol-
lowing ratios: the protection (0, 1) index, quick ratio, real financial cost, EBIT, sales,
debt, Ln(Assets), and foreign exchange.
In the post-devaluation period, the role of financial costs, foreign currency expo-
sure and firm size become more marked as expected. In both pre-devaluation and
post-devaluation periods, the dummy variable for protection has the expected sign
but is not statistically significant. The authors conclude based on this outcome, and
because sectoral regressions reflect contrasts among firms not listed in the stock ex-
change, that higher failure rates for protected firms are concentrated among smaller,
privately held firms.
Although by using probit regression, the authors could evaluate and present the
statistical significance of individual variables, the published statistics (log-likelihood
and the chi-square) do not tell us anything about the classification/misclassification
accuracy among fails and nonfails respectively. In addition, the published results are
in-sample values. Despite the problems with the data, this article is impressive in the
broad sweep of the issues considered in both macroeconomic and microeconomic
terms and in explicitly modeling trade protection and foreign currency exposure. As
we move further into a truly global economy, these variables take on added signifi-
cance in assessing risk.
10.14 BRAZIL. Brazil is an example of an economy where the end result of a series
of economic setbacks would put severe pressure on private enterprises. For example,
tightening of credit for all firms—especially smaller ones—can jeopardize financial
institutions and undermine government efforts to promote economic development.
10 • 32 BUSINESS FAILURE CLASSIFICATION MODELS