Using Prior Knowledge to Forecast Mortality Time Series

Federico Girosi
Rand Corporation

We set the problem of forecasting cause/country/sex/age-specific
mortality in the framework of cross-sectional time series with
covariates. We build a class of priors for the regression coefficients
starting from prior knowledge on mortality. In particular, we show how
to use experts' knowledge on patterns of variation of mortality over
age groups, countries and time to build a rich and detailed class of
smoothness functionals for the expected value of log-mortality. We will
also show that for this class of problems some prior information about
the regularization parameters is available.


Back to Inverse Problems Workshop Series II