Risk analysis and retrospective unbalanced data

• Published in: Revstat Vol. 14; no. 2; p. 157
• Main Authors: Pierri, Francesca, Stanghellini, Elena, Bistoni, Nicolo
• Format: Journal Article
• Language: English
• Published: Instituto Nacional de Estatistica 01.04.2016; Instituto Nacional de Estatística | Statistics Portugal
• Subjects: bankruptcy; case-control studies; Credit ratings; data augmentation; Evaluation; logistic regression; Methods; Regression analysis; Risk assessment; ROSE method; unbalanced data
• ISSN: 1645-6726; 2183-0371
• DOI: 10.57805/revstat.v14i2.184

This paper considers three different techniques applicable in the context of credit scoring when the event under study is rare and therefore we have to cope with unbalanced data. Logistic regression for matched case-control studies, logistic regression for a random balanced data sample and logistic regression for a sample balanced by ROSE (Random OverSampling Examples, Lunardon, Menardi and Torelli, 2014) are tested. We applied the methods to real data: balance sheets indicators of small and medium-sized enterprises and their legal status are considered. The event of interest is the opening of insolvency proceedings of bankruptcy. Key-Words: * bankruptcy; case-control studies; data augmentation; logistic regression; ROSE method; unbalanced data. AMS Subject Classification: * 62J05, 62M20, 62P20, 91G40.