Rank Transformations and the Prediction of Corporate Failure*

This study applies rank transformations to financial ratios to improve the predictive usefulness of standard failure prediction models and suggests that rank- transformed data models show additional improvement in prediction without the added cost of having to predict recession for the companies undergoing testing for potential failure.

Abstract

Abstract Rank transformation of observations has been shown to be useful in linear modeling because the models so constructed are less sensitive to outliers and/or non‐normal distributions than are models constructed using standard methods. In the present study, we apply rank transformations to financial ratios to improve the predictive usefulness of standard failure prediction models. Kane, Richardson, and Graybeal (1996) have shown that failure prediction can be improved by conditioning accounting‐based statistical models on the occurrence of recession. Our results suggest that rank‐ transformed data models show additional improvement in prediction without the added cost of having to predict recession for the companies undergoing testing for potential failure.