Policyholder exercise behavior presents an important risk factor for life insurance companies. Yet, most approaches presented in the academic literature — building on value maximizing strategies akin to the valuation of American options — do not square well with observed prices and exercise patterns.

Following a recent strand of literature, in order to gain insights on what drives policyholder behavior, this paper develops a life-cycle model for variable annuities (VA) with withdrawal guarantees. However, in contrast to these earlier contributions, we explicitly allow for outside savings and investments, which considerably affects the results. Specifically, we find that withdrawal patterns for a standard Guaranteed Minimum Withdrawal Benefit (GMWB) rider are after all primarily motivated by value maximization — but with the important asterisk that the value maximization should be taken out from the policyholders' perspective accounting for individual tax benefits.

On the other hand, preliminary analyses indicate that the discrepancy between the life-cycle model and the subjective value maximization approach increases with the duration of the policy. In particular, for a VA with a Guaranteed Minimum Lifetime Benefit (GMLB) rider, value maximization does not adequately reflect optimal policyholder behavior. We demonstrate that this discrepancy arises from the incompleteness of the financial market due to biometric risks — which become increasingly material as the policyholder’s age increases.

Thus, at a more general level, our findings suggest that while the drivers of optimal policyholder behavior are product-specific, they are systematic in that they directly depend on the “degree of market incompleteness”. We discuss how to assess this “degree of market incompleteness” and derive general prescriptions how to model policyholder behavior in personal savings products.