Title: Large claims modeling Author: Barbora Zuzáková Department: Department of Probability and Mathematical Statistics Supervisor: RNDr. Michal Pešta, Ph.D. Abstract: This thesis discusses a statistical modeling approach based on the extreme value theory to describe the behaviour of large claims of an insurance portfolio. We focus on threshold models which analyze exceedances of a high threshold. This approach has gained in popularity in recent years, as compared with the much older methods based directly on the extreme value distributions. The method is illustated using the group medical claims database recorded over the periods 1997, 1998 and 1999 maintained by the Society of Actuaries. We aim to demonstrate that the proposed model outperforms classical parametric distri- butions and thus enables to estimate high quantiles or the probable maximum loss more precisely. Keywords: threshold models, generalized Pareto distribution, large claims. 1

A company issuing an insurance will provide, in return for a monetary premium, acceptance of the liability to make certain payments to the insured person or company if some beforehand specified event occurs. There will always be a delay between occurrence of this event and actual payment from the insurance company. It is therefore necessary for the company to put aside money for this liability. This money is called the reserve. When a claim is reported, a claim handler will make an estimate of how much the company will have to pay to the claimant. This amount is booked as a liability. This type of reserve is called; "case reserve". When making the estimate, the claim handler has the option of giving the claim a standard reserve or a manual reserve. A standard reserve is a statistically calculated amount based on historical claim costs. This type of reserve is more often used in small claims. A manual reserve is a reserve subjectively decided by the claim handler. This type of reserve is more often used in large claims. This thesis propose a theory to model and calculate an optimal limit above which a claim should be considered large. An application of the method is also applied to some different types of claims.

This bachelor thesis within the field of mathematical statistics aims to study the possibility of predicting specifically large claims from non-life insurance policies with commercial policyholders. This is done through regression analysis, where we seek to develop and evaluate a generalized linear model, GLM. The project is carried out in collaboration with the insurance company If P&C Insurance and most of the research is conducted at their headquarters in Stockholm. The explanatory variables of interest are characteristics associated with the policyholders. Due to the scarcity of large claims in the data set, the prediction is done in two steps. Firstly, logistic regression is used to model the probability of a large claim occurring. Secondly, the magnitude of the large claims is modelled using a generalized linear model with a gamma distribution. Two full models with all characteristics included are constructed and then reduced with computer intensive algorithms. This results in two reduced models, one with two characteristics excluded and one with one characteristic excluded. / Det här kandidatexamensarbetet inom matematisk statistik avser att studera möjligheten att predicera särskilt stora skador från sakförsäkringspolicys med företag som försäkringstagare. Detta görs med regressionsanalys, där vi ämnar att utveckla och bedöma en generaliserad linjär modell, GLM. Projektet utförs i samarbete med försäkringsbolaget If Skadeförsäkring och merparten av undersökningen sker på deras huvudkontor i Stockholm. Förklaringsvariablerna som är av intresse att undersöka är egenskaper associerade med försäkringstagarna. På grund av sällsynthet av storskador i datamängden görs prediktionen i två steg. Först används logistisk regression för att modellera sannolikheten för en storskada att inträffa. Sedan modelleras storskadornas omfattning genom en generaliserad linjär modell med en gammafördelning. Två grundmodeller med alla förklaringsvariabler konstrueras för att sedan reduceras med datorintensiva algoritmer. Det resulterar i två reducerade modeller, med två respektive en kundegenskap utesluten.