Incorporating structural changes in mortality improvements for mortality forecasting

In recent decades, there have been decreasing mortality improvements at younger ages but increasing mortality improvements at older ages in many countries. We propose a modified Lee-Carter method to allow for these structural changes, in which the entire data period is divided into more homogeneous subperiods and a unique set of age-specific parameters is incorporated for each subperiod. We consider a number of methods to project these age patterns into the future. Our results show that the new method can reasonably capture the underlying movements in the age patterns over time and can potentially improve the forecast accuracy of death rates and life expectancies. It is interesting to observe that the highest age sensitivity has been moving gradually to older ages and it is important to take this trend into account in mortality forecasting.