The COVID-19 pandemic led to mortality shocks in many countries worldwide. These shocks complicate the use of stochastic mortality forecasting models. In order to avoid biased projections, it is necessary to identify outliers and control for them. The following resource links cover our materials on robust mortality forecasting.
The resources below cover mortality forecasting only. For analysis of portfolio data affected by shocks like COVID-19, see our page on mortality shocks.
- Univariate robustification for models like Lee-Carter and APC.
- Multivariate robustification for models in the CBD family.
- Robustification of the 2DAP model.
- Robust (log-)likelihoods.
- Identifying borderline unstable ARIMA models.
- Responsiveness to new trajectories in mortality.
How to fit robust mortality-forecasting models using the Projections Toolkit: