< Back to previous page

Project

MICRO: Methods for Individual Claims Reserving Operations

Insurance offers individuals and companies the possibility to manage their risk by transferring future losses to the insurance company in exchange for a deterministic premium. In turn, insurers collect and analyse detailed information on claims and policyholders using predictive analytics to manage these acquired risks. For new policyholders, insurers determine risk-based premiums by predicting their future claim count and size using policyholder characteristics. Once the risk is accepted, the insurer sets aside funds, the so-called reserve, to cover future losses. Nowadays, most insurance companies estimate their reserve based on summary statistics, hereby ignoring the vast majority of the available data in these companies.

This research aims to improve reserve estimates by analysing the detailed information available within the insurance company. In particular, each chapter of this thesis develops a statistical model to predict some of the events (e.g. occurrence, reporting, payment, settlement) registered over the lifetime of a single claim. Through collaboration with several insurers, we aim to develop tools that not only contribute to the literature on claim reserving, but that can also be implemented in practice.

Chapter 2 focuses on the occurrence and reporting of insurance claims. The claim reserve includes all claims that occurred in the past, but due to reporting delays not all of these claims have already been reported. We propose a granular model using daily data to predict the number of incurred, but not yet reported claims. Since less claims get reported during the weekend and on holidays, the reporting delay distribution depends strongly on the occurrence day of the claim. This chapter presents an intuitive approach for modelling the heterogeneity in the reporting delay distribution, which results from these calendar day effects. A simulation study shows that our granular approach has large advantages over traditional methods based on yearly data for portfolios with volatile occurrence processes.

Chapter 3 predicts the future development of reported claims. We present the hierarchical reserving as a modular framework for integrating a claim’s history and claim-specific covariates into the development process. This model can be tailored to any portfolio by adding layers representing the events registered in the portfolio at hand. As a result of this flexibility, many existing reserving models can be rephrased as special cases of the hierarchical reserving model.

Chapter 4 transfers insights from non-life reserving to non-life pricing. Pricing requires a data set with claim counts at the level of individual policies and claim sizes at the level of individual claims. However, an aspect completely ignored by pricing literature is that claim counts and claim sizes are censored due to reporting and settlement delays in the claim development process. We extend the reserving models developed in the previous chapters to remove this censoring from the observed data. The granularity of our model allows it to be applied to both pricing and reserving, hence bridging two key actuarial tasks that have traditionally been discussed in silos. In addition this chapter focuses on the application of reserving methods to reinsurance data, which is more challenging than regular insurance data due to extreme claim sizes, low claim frequencies and long reporting and settlement delays.

The last chapter concludes this work by presenting suggestions for future research in individual reserving. 

Date:27 Sep 2016 →  13 Nov 2020
Keywords:Insurance, IBNR, RBNS, Reserving
Disciplines:Applied economics
Project type:PhD project