Scenario selection with LASSO regression for the valuation of variable annuity portfolios (Q6543145)

For the valuation of a portfolio of life insurance contracts one often has to resort to Monte Carlo simulations because analytic formulae are not available. To evaluate all the contracts would be too time consuming. One therefore chooses a representative sample of contracts to calculate the figures needed. A drawback of this procedure is that information on the contracts not in the valuation sample is lost. The idea of the paper is to reduce the number of basic scenarios and to increase the number of representative samples. Using a first set of representative sample, a Lasso regression is performed on the scenarios to set the weight of some of the scenarios to zero. For a second representative set of contracts, the valuation procedure is performed which needs less time because of the reduced set of scenarios. The method is then applied to the valuation of variable annuity portfolios.