Representative Variable Annuity Policy Selection using Latin Hypercube Sampling

Valuation and risk management of large portfolios of variable annuity policies are a big challenge to insurance companies because pricing a large portfolio of variable annuity policies is a time consuming task. Recently, a method based on clustering and kriging has been proposed to address the computational problem arising from this area. In this method, the value of the portfolio is estimated by the kriging method from the values of the representative variable annuity policies, which are selected by a clustering method. However, thousands of representative policies are required in order to obtain relatively accurate estimations. This paper proposes a Latin hypercube sampling method for selecting representative variable annuity policies. Our test results show that the Latin hypercube sampling method is superior to the clustering method in that only a few hundred representative policies selected by the Latin hypercube sampling method are enough to produce accurate estimations.