Compositional Data Analysis – Coherent Forecasting Mortality Model with Cohort Effect

摘要

In this paper, an extension of the Coherent forecasts of mortality with compositionaldata analysis (CoDa) model of Bergeron-Boucher et al. (2017) to cohort effect isproposed applied to data from six African countries. The process of fitting thismodel starts by adapting the Renshaw and Haberman (2006) to compositional dataanalysis (CODA) as suggested by Bergeron-Boucher et al. (2017). The proposedCoDa-cohort model generally fits the data better than the original cohort model ofRenshaw and Haberman (2006). To get the full CoDa-cohort-coherent model themultiple population factor is included in CoDa-cohort model. Then a comparisonbetween CoDa -coherent and CoDa-cohort-coherent models revealed that they havesimilar accuracy for the selected countries in West Africa but not for countries inEast Africa based on Aitchinson distance (AD). But for merged populations likemale and female, the new model, CoDa-cohort-coherent, has generally better fitsfor Kenya mortality data.