Mortality Forecasting

The 2000 birth cohort required forecasts for mortality from ages 23 through 110. Forecast validation showed modest median deviations and a maximum deviation of 3.6%. The study forecast future mortality because many cohorts had not been fully observed through death by 2022. The study applied the Lee-Carter mortality forecasting model separately for each state, region, and the national level. A parsimonious forecasting approach would keep the Gompertz slope stable while allowing mortality levels and short-term period variation to change. Mortality forecasting models may not need unlimited long-run drift in the slope of the log-mortality curve. The framework can help separate mortality level and exposure-history differences from differences in the pace of aging.