A Comparison of Classification/Regression Trees and Logistic Regression in Failure Models

The use of non-parametric statistical methods, the development of models geared towards the homogeneous characteristics of corporate sub-populations, and the introduction of non-financial variables, are three main issues analysed in this paper. This study compares the predictive performance of a non-parametric methodology, namelyClassification/Regression Trees (CART), against traditional logistic regression (LR) by employing a vast set of matched-pair accounts of the smallest enterprises, known as micro-entities,from the United Kingdom for the period 1999 to 2008 that includes financial, non-financial, and macroeconomic variables. Our findings show that CART outperforms the standard approach in the literature, LR.