A comparison of mortality forecasting models

Tapas Mishra, International Institute for Applied Systems Analysis
Monika S Sawhney, Tulane University

Since the famous Lee-Carter (1992) method for forecasting mortality was published a wide-range of models have been floated based on both parametric and non-parametric family and in Bayesian and Non-Bayesian domains. While the basic objective of any forecasting method in this aspect remains the same, its treatment of uncertainty widely differs. Most of the recent developments have occurred in the classical (i.e., non-Bayesian) domain whereas some recent research show the power of Bayesian models against classical methods. In case of the latter, variants of Auto regressive moving average (ARMA) class have been extensively utilized, whereas some research also point to the use of non-parametric or even semi-parametric methods. In view of enormous socio-economic-demographic significance of mortality forecasts, we intend to provide a comparison of various forecasting methods using both classical and Bayesian tradition. A new forecasting method would then be proposed based on our evaluation of different models.

  See extended abstract

Presented in Poster Session 5: Contexts