2023 Australasian Actuarial Education and Research Symposium

Gayani Thalagoda

University of New South Wales

Shapley decomposition-based selection of representative contracts for variable annuity portfolio valuation

This is joint work with Katja Hanewald, Andrés M. Villegas, Jonathan Ziveyi

The paper presents a Shapley decomposition-based method to enhance explainability in the selection of cluster representatives for valuing variable annuity portfolios. While existing clustering-oriented data mining frameworks offer notable reductions in computational time, the selection of cluster representatives using these methods is independent of the risk measure being calculated. Policies with seemingly similar risk characteristics in the feature space may end up generating significantly different cash flows, diminishing their practical appeal for principle-based calculations. As a solution, this study proposes an algorithm that forms clusters based on Shapley decompositions. The method involves decomposing the overall risk of a contract into clearly separated contributions from each risk driver using a Shapley value-based decomposition. This decomposition allows for a structured and meaningful representation of the policy data, which is then used for selecting the cluster representatives. The proposed method can assist users in explaining the reasoning behind selection of a policy as a cluster representative. Furthermore, the proposed method aligns with the grouping requirements of VM-21: Requirements for Principle-Based Reserves for Variable Annuities, which necessitate representative policies to be selected in a manner that accurately reflects characteristics and criteria with a material impact on the calculated risk measure.