Comparing the predictive performance of the logit and discriminant analysis models in bankruptcy prediction

Type of content
Theses / Dissertations
Publisher's DOI/URI
Thesis discipline
Accountancy
Degree name
Master of Commerce
Publisher
University of Canterbury
Journal Title
Journal ISSN
Volume Title
Language
English
Date
2018
Authors
Ono, Mari
Abstract

This research assesses the sensitivity of ratio analysis in bankruptcy prediction. I challenge the usually applied criterion in bankruptcy analysis where a model outperforms another model based on their classification performance (prediction) of firms into bankrupt and non-bankrupt categories. The commonly applied decision criterion is based on Type 1 and Type 2 errors or the aggregate predictive ability of a model. However, I claim that, say, if these metrics show identical results for two models, it is not clear by any stretch that they have the same predictive ability. Rather I contest that looking at the classification patterns of each individual firm, one may observe that indeed both models classify rightly and wrongly all firms in a testing sample; or, they do not, in which case a data-driven result has been obtained.

I have chosen to demonstrate this novel idea by means of the classification performance of the multivariate discriminant analysis (MDA) and Logit models. These two models were introduced by Altman (1968) and Ohlson (1980), respectively, and are probably the most established bankruptcy prediction tools in the literature and in business applications. In doing so, I heavily lean on Begley et al. (1996) who compare the two models using data from the 1990ies. From this starting point, I further apply criteria that allow for a fairer comparison of the two models, because testing two models about the predictive ability to classify firms into groups seems unreliable when each model uses different variables and where for neither of the two models any specification test results are considered.

My research question therefore asks whether the reported results of the MDA and Logit models in the literature with respect to bankruptcy prediction do hold when firm-specific classification patterns are considered. I provide various negative results to this question. This work and the discussed future direction that may be taken based on my work have to potential to redirect the bankruptcy literature to more useful assessments of model performance comparisons.

Description
Citation
Keywords
Ngā upoko tukutuku/Māori subject headings
ANZSRC fields of research
Rights
All Rights Reserved