Smoothing Mortality Rates Using Random Forests

Torsten Sauer , University of Rostock
Roland Rau, University of Rostock

To smooth sparse mortality age schedules, existing approaches use parametric, penalized non- parametric, relative and Bayesian approaches. Statistical learning has become important in various scientific disciplines to solve complex problems. The question arises if it can also be used to smooth mortality age schedules. We introduce a new algorithm based on Random Forests, a non-parametric statistical learning approach. The algorithm is trained on data from the Human Mortality Database about the shape of smooth mortality schedules. The trained models are then used to predict smooth age trajectories on sparse data. To test the method we apply cross validation with simulations of sparse data. It is found that the approach can smooth moderate sparse patterns comparably well to TOPALS. The algorithm introduces a new way of smoothing and its promising results makes it also considerable for age-specific fertility rates.

See extended abstract

 Presented in Session 29. Estimation Methods