Insurer’s Insolvency Prediction Using Random Forest Classification
Series
Working paper
Type
working paper
Date Issued
2015-02-22
Author(s)
Abstract (De)
This paper uses a modification of the Random Forest classification algorithm to predict insolvency of insurers. RF orders companies according to their propensity to default. We show that RF methodology delivers higher quality of prediction compared to other existing methods. In addition, RF classification can be used to gather further insights about the fragile companies. It ranks the explanatory variables in the order of their ability to predict insolvency. Also it is used to describe the relationship between the propensity to default and the individual characteristics of an insurer. We show that many of these relationships are highly non-linear.
Language
English
HSG Classification
contribution to scientific community
Subject(s)
Division(s)
Eprints ID
260655