Spatio-Temporal Extensions of the Lee-Carter Model

Peter Vekas, Corvinus University of Budapest 1093 Budapest Fovam ter 8.
Gabor Szentkereszti, Corvinus University of Budapest

The overwhelming majority of researchers, actuaries and demographers use standard time series analysis techniques to project time-varying parameters of popular mortality forecasting methods such as the Lee–Carter and Li–Lee models. However, spatial dependence can be as significant as temporal autocorrelation in these time series, and the underlying panel structure of the data is often neglected. We draw from techniques in panel and spatial econometrics, including ordinary and spatial dynamic panel linear models, spatial eigenvector filters and spatial autoregressive models to capture this kind of dependence and yield more accurate projections. We examine our methods on mortality data from a large set of 24 countries and point out which methods work best for each country, separately for both the Lee–Carter and Li–Lee models. We demonstrate that our proposed techniques outperform conventional methods in most populations studied. The improved forecasts could be particularly useful in life insurance and pensions and in insurance contexts involving longevity risk.

Keywords: Mortality and Longevity, Spatial Demography, Econometrics , Data and Methods

See paper.