Bayesian Log-Bilinear Mortality Projection with a Random Walk with Drift

This paper extends the widely used Lee Carter (LC) model (Lee & Carter, 1992) for mortality projection. We suggest a random walk with drift to model the time parameter of the Bayesian extension of the LC model suggested in Czado et al. (2005). In a validation-based examination, the proposed random walk model produces smaller prediction errors compared to the autoregressive model in Czado et al. (2005).The random walk with drift model is valid only if the linear trend remains constant over time. We further extend the random-walk model to allow for change-points in the time parameter. In a validation-based examination, the proposed change-points model produces smaller prediction errors compared to a model with constant drift parameter