Mortality Projection: Curve Fitting And Linear Regression

Abstract

This study was aimed at predicting future mortality rates given two main scenarios. The first is in the case where there is a body of mortality tables from the past and the second is in the case where such a database does not exist. In the first case, linear regression was used to estimate mortality of specific age groups for a specific future year. In this case the specified future year was from 2000 to 2009 although similar computations can be carried out for age specific and single year life tables. To perform these linear regression two transformations were considered to the existing q type mortality rates: the log linear transformation and the logit transformation For the second case, it was assumed that the only data that existed was that of actual deaths and exposed to risk. Thus, for adequate mortality projections to be done a Makeham curve was fitted in one case and a cubic spline graduation was done for the second case. Both the Makeham and the cubic spline methods were observed to have a good fit to the data, the latter providing a better fit than the former. However the linear regression methods were observed to give an almost constant difference at every age and would probably not be the best methods of forecasting especially for very long time periods. The advantage of the linear regression method was that it was seen to keep the original shape of the graduation