Actuarial Notes

Mortality Improvements Research: Report to the Canadian Institute of Actuaries' Project Oversight Group

By Kai Kaufhold, Martina Brück and Claus Neidhardt · Canadian Institute of Actuaries · 2024 · Document 224043

Study guide: Exam RET-301

A research report, dated April 2024, on mortality improvements (MI) in Canada, carried out by Ad Res Advanced Reinsurance Services GmbH with Koblenz University of Applied Sciences on behalf of the Canadian Institute of Actuaries. Its purpose is to provide a basis for an updated set of mortality improvement assumptions for use by actuaries across all practices. It recommends best-estimate improvement rates from an adapted version of the CMI Method, and states that sponsored research of this kind does not constitute guidance.

On the syllabus

  • Exam RET-301 — Outcomes 1a, 1b, 1c (Assumptions); the whole report (CIA Report, Apr 2024).

1 Executive Summary

  • The recommendation adapts the CMI Method used by the UK Continuous Mortality Investigation, which decomposes observed improvement rates into age, period and cohort components and relies on expert input for their long-term behaviour.
  • The recommended core long-term improvement rate is 1.3% for both males and females, with a standard error of around 0.5%; for ages 20 to 40 it is 1.0%.
  • Of 14 stochastic mortality models tested, the Age-Period-Cohort (APC) model and the integrated Age-Period-Cohort (APCI) model were selected.
  • The data are Human Mortality Database figures for the general Canadian population, with a core age range of 40 to 90 and experience from 1980 to 2019.
  • Using calendar years 2020 or 2021 as a jump-off point would lead to distortions, in the authors' review of the effect of COVID-19.

2 Data

  • Choice of Data Source
  • Features of the Data

3 Age-Period-Cohort Model and APCI Model

  • Parameter Estimation
    • Identifiability
    • Sparse cohorts
    • Smooth parameter estimation
  • Time series projections
  • Uncertainty of stochastic projection and parameter estimation

4 Adapted CMI Method

  • CMI Method
    • Deterministic projection
    • Polynomial extrapolation
    • Deriving initial MI rates from the APCI model
  • Equating CMI Method to Stochastic Forecasts

5 Extrapolation for Advanced Ages

6 Implementation, results and discussion

  • Parameter Estimates
  • Alternative ARIMA Model for APC κ_y
    • Discussion of time series process
  • Stochastic and Deterministic Forecasts
  • Smooth Kappa Implementation for Future MI
  • Impact Analysis
  • Choice of Study Period
  • Choice of Convergence Period and Projection Period

7 Outlook

  • The report recommends a scale from the adapted CMI Method with a deterministic projection to an age-independent, unisex long-term improvement rate after 30 years.

8 References

Appendix

  • Qualitative and Quantitative Model Selection Criteria
  • List of stochastic mortality models investigated
  • Recommendation for Ages 20-39
  • List of Abbreviations

Sources