Beyond annual data: Mortality forecasting with mixed frequency data

Existing mortality models typically rely on annual data, with forecasts for a given year based on information available up to the end of the previous year. However, technological advances have enabled the collection and production of weekly and monthly death data, offering new opportunities to improve forecasting. Using only annual data overlooks the full range of available information. In this paper, we propose mixed frequency sampling (MIDAS) models to integrate monthly death counts (high frequency) with annual mortality rates (low frequency), enabling improved short-term prediction of annual mortality. Extending economic applications of MIDAS, which typically predict single variables such as GDP growth, our MIDAS framework accounts for age dependence unique to age-specific mortality modeling. We also evaluate different weighting functions, a core element in MIDAS that determines the relative importance of high-frequency data at different lags, and identify suitable weighting functions for mortality forecasting. Using U.S. mortality data, we demonstrate that our approach significantly improves prediction accuracy compared to models relying solely on annual data for short-term forecasting. These findings highlight the potential of MIDAS models as a useful tool for accurate and timely mortality forecasts.