Improving HMD Mortality Estimates with HFD Fertility Data

Abstract

This article aims to improve mortality estimates using fertility data. Estimating the population exposed to risk, such as in the Human Mortality Database (HMD), can suffer from errors for cohorts born in years in which births fluctuate unevenly over the year. When comparing period and cohort mortality tables, we highlight the presence of anomalies in the period tables in the form of isolated cohort effects. Our investigation of the HMD methodology shows that it assumes a uniform distribution of births that is specific to the period tables, which is likely to lead to an asymmetry with the cohort tables. Building on the “Phantoms Never Die” study of Cairns et al. regarding the construction of a “data quality indicator,” we utilize the Human Fertility Database (HFD), which is the perfect counterpart to the HMD in terms of fertility. The indicator is then used to construct corrected period mortality tables for several countries, which are then analyzed from both historical and prospective points of view. The analysis has implications for the reduction of volatility of mortality improvement rates, the use of cohort parameters in stochastic mortality models, and the improved fit of corrected tables by classical mortality models.

ACKNOWLEDGMENTS

The author is grateful to Laurent Devineau for fruitful research directions, as well as for input, discussions, and remarks on this work. The author also thanks his collegues at Milliman, in particular Sophie Decupére and Mohamed Talfi, for several interesting discussions.

Discussions on this article can be submitted until October 1, 2020. The authors reserve the right to reply to any discussion. Please see the Instructions for Authors found online at http://www.tandfonline.com/uaaj for submission instructions.

Notes

1 Both sides are multiplied by –1 to get the (positive) number of deaths.

2 Number of children already born.

3 We do not consider preliminary release HFD data because they have not been fully processed and checked.