Logistic discrimination versus tree-based models: A case study of Australian companies' use of derivative markets
by
Steven Stern
ANU
Coauthors: Richard Heaney (ANU)

In many fields, logistic discrimination is a standard technique employed to distinguish between two distinct groups within a given population based on a set of multidimensional covariate information regarding observed elements of that population. However, it seems rare that any diagnostic procedures are employed to examine whether the very specific model structures implied by this analysis are reasonable. We examine a dataset regarding the use of derivative markets by Australian companies as hedges against future company value. We first employ the standard logistic methods and then demonstrate that this model seems to be inadequate, at least in its simplest form. Finally, we employ a more exploratory tree-based analysis and discuss its results and implications. In particular, some modern computer-intensive methods for refining simple tree-based models are briefly discussed.

Date received: April 29, 2002