Validating Bayesian MCMC models within stochastic loss reserving

Specialeforsvar ved Martin Fiil Laugesen 

Titel:  Validating Bayesian MCMC Models Within Stochastic Loss Reserving

 Abstract: This paper will investigate the models proposed by Glenn Meyers in the article "Stochastic Loss Reserving Using MCMC Models". The theory behind the MCMC algorithms will be discussed, Meyers models will be described in great details, and an implementation of the methodology will be outlined. The implementation proposed outperforms Meyers implementation in accuracy. 2 additional models will be introduced correcting correlation in the original models. All models are tested on a new dataset. The incurred models estimate the distribution of the reserve ultimate almost perfectly. The paid models fail to give the correct estimations of the distribution of the reserve ultimate. The overall capabilities of modelling within the Markov Chain Monte Carlo framework shows potential to explore further models with a high level of complexity and huge numbers of parameters

Vejleder: Jostein Paulsen
Censor:   Mette M Havning