
Linear discriminant analysis was introduced by Fisher (1936) and developed in the
1960s by Beaver (1967) and Altman (1968), whose Z-score substantially contributed to
the diffusion of credit-scoring models. The underlying idea can be explained as follows.
Let us consider a population of n fi rms for which m different variables (e.g., balance
sheet ratios) are observable. Some of the n fi rms are “good,” i.e., fi rms that did not default
within a certain period (usually one year) from the time the m variables were observed,
while the other, “bad,” fi rms defaulted within the same time frame. Linear discriminant
analysis tries to identify a scori