Collective Reserving using Individual Claims Data

Abstract

The aim of this paper is to operationalize claims reserving based on general insurance individual claims data. We design a modeling architecture that is based on six different neural networks. Each network is a separate module that serves a certain modeling purpose. We apply our architecture to individual claims data and predict their settlement processes on a monthly time grid. A proof of concept is provided by benchmarking the resulting claims reserves with the ones received from the classical chain-ladder method which uses much coarser (aggregated) data.

Keywords: claims reserving, general insurance, individual claims data, micro-level reserving, neural networks, IBNR claims, RBNS claims, chain-ladder method, over-dispersed Poisson model

JEL Classification: G22, C22, C25, C45, C51, C53