Forecasting cause-specific mortality using time series methods

Age profiles of cause-specific mortality for five leading causes of death are represented by parameterized mortality schedules. Such schedules are estimated for males and females over the period 1960-1985. The resulting time series of parameters are modeled as ARIMA processes, and parameter forecasts from these models yield forecasts of the complete age profile of mortality-by-cause. Forecasted profiles for most causes show satisfactory mean absolute percent errors and Theil statistics, with improvements relative to trend extrapolation methods at the longer forecast horizions.