Decision Trees and Financial Variables

Abstract

A decision tree program for forecasting stock performance is applied to Compustat's Global financial statement data augmented with International Monetary Fund data. The hypothesis is that certain Compustat variables will be most used by the decision tree program and will provide insight as to how to make investing decisions. Surprisingly, the authors' experiments show that the most frequently used variables come from the International Monetary Fund and that variables provided exclusively for Financial Industry stocks were not useful for forecasting financial stock performance. These experiments might be part of a constellation of such experiments that help people map financial forecasting problems to the variables most useful for solving those problems. The research shows the value of using decision tree methodologies as applied to finance.