Unmodelled aspects of longevity could invalidate insurer's predictions for policy pricing, reinsurer RGA has said.
Actuaries need to make predictive models when developing financial products in order to forecast mortality rates, inflation and investment returns that will be available in future.
A longevity conference held at Warwick University, heard actuaries Greg Becker and Garth Lane, both of RGA, talk on the role of heterogeneity (the quality of being diverse), and how unmodelled aspects could invalidate a model's predictions.
They illustrated their message by considering a scenario in which two models could be used which both produced similar results for many years before diverging due to the underlying heterogeneity between sub groups of lives, such as participation in sport or fast food consumption, and discussed how to choose between models.
Becker said: "If the data which shows which model is right will not be available for many years - say annuitant survival data - what uncertainties should actuaries consider when making their predictions?
"It is these kinds of considerations that affect the pricing of long term products like annuities and life insurance, and when setting up reserves for long term commitments."
The pair concluded that financial institutions should take into account underlying sub populations and the causes of death that act on them.
Steps could be taken to realistically reflect the level of improvement to mortality and also associated uncertainty.
"If actuarial models assume that everyone is the same, they may lead to surprises in future, surprises that could have implications for some policyholders and annuitants," said Lane.