In recent decades, much attention has been given worldwide to mortality pattern analysis and forecasting. Projections of mortality rates are of primary importance for both governments and private companies in a number of fields, such as private life and health insurance, pension funds and long-term care homes. Projection of mortality rates requires stochastic models for a proper estimation of the future average life expectancy dynamics, as well as longevity risk, in a systematic way.
This thesis empirically compared two procedures of estimating the multi-population mortality model parameters, i.e. the frequentist approach and Bayesian inference. Additionally, a Poisson mixture modification of the considered models was studied, and the obtained results were compared to the basic versions of the corresponding models.
The data for five European countries from the Human Mortality Database from 1951 to 2000 were used to estimate the model parameters. The data from 2001 to 2010 were applied for backtesting the forecast quality of the considered models. The log-mortality rates obtained by all considered models were compared based on the unknown parameter estimates, the goodness of fit, and their prediction performance.
«
In recent decades, much attention has been given worldwide to mortality pattern analysis and forecasting. Projections of mortality rates are of primary importance for both governments and private companies in a number of fields, such as private life and health insurance, pension funds and long-term care homes. Projection of mortality rates requires stochastic models for a proper estimation of the future average life expectancy dynamics, as well as longevity risk, in a systematic way.
This thesi...
»