Examining structural shifts in mortality using the Lee-Carter method

We present an extension of the Lee-Carter method of modeling mortality to examine structural shifts in trajectories of mortality. Austrian data consisting of 53 years of single-age mortality rates are subdivided into 30 24-year submatrices. Using singular value decomposition, the submatrices are decomposed into three component submatrices: 1) the multiple realizations of the index of mortality to which each respective age-specific death rate is linearly related; 2) the average shape across age of the log of mortality schedules; 3) the sensitivity of the log of mortality at each age to variations in the elements of the index of mortality. We refer to these latter submatrices to locate structural changes in mortality patterns. A comparison of the observed and estimated life expectancy indicates that the extended Lee-Carter method is superior to the original Lee-Carter method, particularly so for life expectancies at higher ages. We conclude by projecting life expectancy up to 2050, applying the Lee-Carter method to the whole time series (1947-1999) and comparing it to an application of the Lee-Carter method to the latest subsample (1976-1999). (AUTHORS)