Deakin University
Browse

File(s) under permanent embargo

Fast valuation of large portfolios of variable annuities via transfer learning

conference contribution
posted on 2019-01-01, 00:00 authored by X Cheng, Wei LuoWei Luo, G Gan, Gang LiGang Li
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.

History

Volume

11672

Pagination

716-728

Location

Cuvu, Fiji

Start date

2019-08-26

End date

2019-08-30

ISSN

0302-9743

ISBN-13

9783030298937

Language

eng

Publication classification

E1 Full written paper - refereed

Editor/Contributor(s)

Nayak A, Sharma A

Title of proceedings

PRICAI 2019: Trends in Artificial Intelligence

Event

Pacific Rim International Conference of Artificial Intelligence (2019 : Cuvu, Fiji)

Publisher

Springer

Place of publication

Berlin, Germany

Series

Lecture Notes in Computer Science

Usage metrics

    Research Publications

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC