A Markovian Aging Process Forecasting Model: Predicting U.S. Mortality

Abstract

In this study, we provide a novel finite-state Markov model for predicting death rates. The Markovian physiological age, which forms the basis of this model, is represented by the states in the underlying continuous-time Markov chain. This model forecasts mortality rates for the Markovian physiological age, for which mortality rates for calendar ages can be easily computed. A set of data from the U.S. population is used to calibrate the model. The data collection includes individuals aged 30 to 108 and covers the years 1970 to 2019. We train the model using data from 1970 to 2014 and then test it using data from 2015 to 2019. Based on metrics utilized for training and test datasets, the suggested model outperforms the models of both Lee and Carter (1992) and Renshaw and Haberman (2006). It is noteworthy that this method uses fewer model parameters than the comparable models. Forecasts of mortality rates and life expectancy are made using the findings. According to the findings, a 30-year-old’s life expectancy in 2040, 2060, and 2080 will be 55, 59, and 63 years, respectively, which is longer than the basic Lee-Carter model predicted. The model in this work, unlike the Lee Carter model, is identifiable.