ABSTRACT
We use the example of health expenditure in the Netherlands by age, sex and time, to show how the framework of Part III can be applied to nondemographic events such as tax payments, crimes, and hospital admissions.
We first examine the data. We set up four scenarios based on a conventional projection model. The baseline scenario assumes that expenditureper capita within each age-sex group continues to grow at the average historical growth rate. In the “high” or “low” scenario, each age-sex group’s future growth rate equals the corresponding historical growth rate plus or minus 2 percentage points. In the “fixed” scenario, future growth rates are 0. We show that different scenarios give dramatically different results regarding total expenditure and expenditure per capita, but virtually identical results regarding the percentage of total expenditure that is spent on people aged 85+. A second problem with the scenario-based approach is that the likelihoods of the scenarios are unclear.
We set up a Bayesian version of the projection model, check the model with replicate data technique, and construct expenditure projections. We show that the Bayesian projections overcome the limitations of the conventional projections, and provide measures of uncertainty that have policy relevance.
