A note on performance measures for failure prediction models

Since decades, the topic of business failure prediction has been an important research area for both academics and practitioners. Bankruptcy prediction involves the classification of firms in a failing and a non-failing group3. Generally, this classification is based on (1) a prediction model that attributes a ‘score’ to each firm in the data set and (2) a certain cutoff point. To evaluate the classification results, several performance measures can be used. This note outlines these measures and illustrates the connections between them with numerical examples. This may help the reader to better understand (and possibly use) these classification measures.