Title: Underwriting fraud prediction based on conditional density estimations
Authors: Felix Vandervorst - Allianz Benelux (Belgium) [presenting]
Abstract: Underwriting premium fraud is the risk of adverse data misrepresentation committed with the intent to benefit from an undue lower premium. We propose a novel approach to quantify underwriting premium fraud risk at application time for an insurance pricing model under identifiability of the conditional distribution assumption and availability of non-misrepresented historical quote data. The approach does not require historical premium fraud labels and adapts to change in pricing policy, unlike most supervised and unsupervised approaches to fraud detection. Moreover, our approach can be used to detect outliers next to predicting underwriting fraud and is extensible to multivariate data misrepresentation. We illustrate the approach with motor insurance underwriting data, where the driver identity may be misrepresented to benefit from an undue lower premium.