Modeling Longevity and Disability with Generalized Autoregressive Score Models

Multi-state models are widely used in the actuarial modeling of mortality and disability. To account for longevity and time variations in transition rates, the existing literature relies on (parameter-driven) state-space models. Despite its flexibility, the state-space model is computationally intensive as it requires the evaluation of a high-dimensional integral to compute the likelihood function. This research proposes an alternative (observation-driven) multi-state model with time-varying transition rates for mortality and disability modeling based on the generalized autoregressive score (GAS) models. The proposed model is still very flexible and is superior to existing methods in terms of computational efficiency. We illustrate our model using the Health and Retirement Study (HRS) data by discussing estimation and simulation issues