A0688
Title: Application of artificial neural networks to variable annuities
Authors: Longhai Li - University of Saskatchewan (Canada) [presenting]
Abstract: The brute-force Monte Carlo (BFMC) simulation is a straightforward method that has long been used to evaluate the risk of financial portfolios. Although the BFMC method works reasonably well for the risk management purposes, it usually involves very intensive computation, due to the nested simulation structure and the needs to simulate a large number of inner loops. To reduce the computation cost, the least-squares Monte Carlo (LSMC) method is used to simulate the liability values. However, the LSMC methods do not work very well in estimating the performance of hedging strategies and the Greeks of variable annuities, and additional research needs to be done to improve the performance of the LSMC method. Potential improvement can be achieved by applying the neural network curve fitting method to the fitting of the proxy functions of liabilities and Greeks, so that better proxy functions can be obtained by allowing more curve fitting flexibilities.
