Authors: Benjamin Furniss, Britton Borget, John Sanders
Mentors: Patrick Ling
Insitution: Utah Valley University
Mortality model is the underlying model used by life actuaries to price life policies, set reserve amounts, and compute policy values. A mortality model investigates how mortality rates evolve over time. Current insurance law in many states (including Utah) suggest the use of Scale AA (or a similar model) in projecting future mortality rates, which is a special case of autoregression time series model. This model is flawed because it is built on the assumption that (1) there is no ARCH effect in the central death rates data, and (2) there is no unit root in the time series of mortality index. These assumptions are questionable. No wonder why state insurance laws (including Utah state insurance law) are recently revised in recognition of discrepancy between model predicted mortality rates and actual mortality rates. Recent published literatures indicate that the second assumption is questionable, as some statistical tests suggest that there is some near unit root in the mortality model. In this talk we want to argue that ARCH effect is present in the mortality data, so there is need to adopt a time series model that incorporates heteroskedasticity in the mortality data. We will later propose a GARCH model for better predicting future mortality rates – a key task life actuaries conduct, for it is important for life actuaries to predict what will happen over the next few decades of policy term.