Prediction of Mortality Rates in the Presence of Missing Values.

A time series model based on multivariate power-normal distribution has been applied in the past literature on the United States (US) mortality data from the years 1933 to 2000 to forecast the future age-specific mortality rates of the years 2001 to 2010. In this paper, we show that the method based on multivariate power-normal distribution can still be used for an incomplete US mortality dataset that contains some missing values. The prediction intervals based on this incomplete training data are found to still have good ability of covering the observed future mortality rates although the interval lengths may become wider for long-range prediction. [ABSTRACT FROM AUTHOR]