Vinnere av aktuarprisen 2026

Aktuarprisen for 2026 tildeles Skjalg Klausen og Silje Lunde Husebø.

Skjalg Klausen fra Universitet i Oslo og jobber i Gabler

For masteroppgave med tittel: Probability-Free Option Pricing via Rough Path Theory and Applications to Insurance,

Sammendrag:

Masteroppgaven bygger på et etablert, sannsynlighetsfritt rammeverk, basert på rough-path-teori, for prising og hedging av opsjoner, med anvendelser til reservering og prising av livsforsikring med investeringsvalg.I motsetning til klassisk matematisk finans, som forutsetter en sannsynlighetsmodell for aktivapriser og dermed introduserer modellrisiko, formuleres analysene sti-vis uten sannsynlighetsteori. Tilnærmingen kan redusere avhengigheten av modellantakelser og styrke robustheten i aktuariell praksis.Vi gjennomgår prinsipper og metoder i rough-path-teori og sannsynlighetsfri prising, og diskuterer anvendelser for unit-linked produkter. I tillegg gis en grundig innføring i klassisk matematisk finans, livsforsikringsmatematikk samt nødvendig analytisk og stokastisk bakgrunnsmateriale.

Silje Lunde Husebø fra Universitet i Bergen og jobber i Gjensidige

For masteroppgave med tittel: Claim frequency estimation - improving model stability using penalized regression in a GLM framework

Sammendrag:

Generalized Linear Models (GLM) are widely used in the pricing and modelling of non-life insurance. The insurance market is a highly competitive market and is regulated and closely supervised to ensure that the solvency requirements are met. Therefore, it is crucial to have a stable pricing model that accurately reflects the risk in the insurance portfolio. GLMs are generally known to be stable and robust but can in some cases cause unstable results when updating models with new data over time. LASSO and Ridge regression are alternative methods that are known to reduce variance, which is closely related to stability, by shrinking the parameters.

In this thesis, two main objectives are analysed when modelling claim frequency. The first objective is to analyse how the hyperparameter lambda of the LASSO and the Ridge regression impacts their stability and performance. The second objective is to develop new penalized regression models with improved stability based on the GLM framework, explicitly taking into account a stability criterion through the penalty term, while maintaining predictive accuracy. For this, three models are proposed and compared with the performance and stability of the GLM, LASSO and Ridge regression. The models are tested on several datasets, both simulated and R-datasets. The general results demonstrate that the proposed Stability-adjusted Penalized regression models improve both the stability and the overall performance compared to the GLM, LASSO and Ridge regression.

Den Norske Aktuarforening gratulerer!