The sensitivity of the Scaled Model of Error with respect to the choice of the correlation parameters: A Simulation Study
Rebecca Graziani and Nico Keilman
The Scaled Model of Error has gained considerable popularity during the past ten years as a device for computing probabilistic population forecasts of the cohort-component type. In this report we investigate how sensitive probabilistic population forecasts produced by means of the Scaled Model of Error are for small changes in the correlation parameters. We consider changes in the correlation of the age-specific fertility forecast error increments across time and age, and changes in the
correlation of the age-specific mortality forecast error increments across time, age and sex. Next we analyse the impact of such changes on the forecasts of the Total Fertility Rate and of the Male and Female Life Expectancies respectively. For age specific fertility we find that the correlation across ages has only limited impact on the uncertainty in the Total Fertility Rate. As a consequence, annual numbers of births will be little affected. The autocorrelation in error increments is an important parameter, in particular in the long run. Also, the autocorrelation in error increments for age specific mortality is important. It has a large effect on long run uncertainty in life expectancy values, and hence on the uncertainty around the elderly population in the future. In empirical applications of the Scaled Model of Error, one should give due attention to a correct estimation of these two parameters.