Variable annuities are important financial products that result in 100 billion sales in 2018. These products contain complex guarantees that are computationally expensive to value, and insurance companies are turning to machine learning for the valuation of large portfolios of variable annuity policies. Although earlier studies, exemplified by the regression modelling approach, have shown promising results, the valuation accuracy is unsatisfying. In this paper, we show that one main cause for the poor valuation accuracy is the inefficient selection of representative policies. To overcome this problem, we propose a novel transfer-learning based portfolio valuation framework. The framework first builds a backbone deep neural network using historical Monte Carlo simulation results. The backbone network provides a valuation-driven representation for selecting the policies that best represent a large portfolio. Furthermore, the transferred network provides a way to adaptively extrapolate from these representative policies to the remaining policies in the portfolio. By overcoming a major difficulty faced by the popular Kriging model, the need of matrix inversion, the transferred network can handle a large number of representative policies to sufficiently cover a diverse portfolio.
CITATION STYLE
Cheng, X., Luo, W., Gan, G., & Li, G. (2019). Fast Valuation of Large Portfolios of Variable Annuities via Transfer Learning. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11672 LNAI, pp. 716–728). Springer Verlag. https://doi.org/10.1007/978-3-030-29894-4_57
Mendeley helps you to discover research relevant for your work.