We use a stochastic age-period-cohort mortality model to analyse US data for years 1989-2015 and ages, separated by gender, educational attainment, and cause of death. The paper focuses, in particular, on the fitted cohort effect for each sub-population and cause of death with two key findings. First, causes of death with a strong or distinctively-shaped cohort effect are also causes of death with significant, controllable risk factors, and that the fitted cohort effect gives us insight into the underlying prevalence of specific risk factors (such as smoking prevalence). Second, although each sub-population and cause of death has its own distinctive model fit, there are sufficient similarities between cohort effects to allow us to postulate that there is a relatively small number of underlying controllable risk factors that drive these cohort effects. The analysis then gives us insight into the modelled cohort effect for all-cause mortality.
|Journal||Insurance: Mathematics and Economics|
|Publication status||Accepted/In press - 26 Feb 2021|