Oscar Lindberg: Return period estimation for wind storm losses in Norway
MSc Thesis Presentation
Time: Tue 2021-06-08 09.00
Location: Zoom, meeting ID: 646 0130 8139
Respondent: Oscar Lindberg
Supervisor: Filip Lindskog
The insurance risk for natural catastrophes is highly determined by extreme loss events occurring with low frequency. For insurance providers, this makes inference on individual catastrophe events a crucial part of inference. In this thesis, we analyze historical data for wind storms in Norway in order to model the probability of storm events with extreme losses. To do this, we use a Peaks over threshold model, in which the excess losses for each storm event above a selected cost threshold follows a Generalized Pareto distribution. For this distribution, we compare maximum likelihood estimation to a Bayesian model. The prior for the Bayesian model is based on a separate catastrophe model, which generates potential storm events to assess the probability of catastrophe loss scenarios. Estimation uncertainty is evaluated using profile likelihood methods, a Bootstrap analysis, and evaluation of the the posterior distribution for the Bayesian model.