Insurance Supervisory Authority of Finland,

In this paper a Bayesian approach is utilized to analyze the role of the underlying asset and interest rate model in the market consistent valuation of life insurance policies. The focus is on a novel application of advanced theoretical and computational methods. A guaranteed participating contract embedding an American-style option is considered. This option is valued using the regression method. We exploit the flexibility inborn in Markov Chain Monte Carlo methods in order to deal with a fairly realistic valuation framework. The Bayesian approach enables us to address model and parameter error issues. Our empirical results support the use of elaborated instead of stylized models for asset dynamics in practical applications. Furthermore, it appears that the choice of model and initial values is essential for risk management