Functional Data Analysis in Forecasting Serbian Fertility

A new approach, combining functional data analysis and principal components decomposition in order to forecasting demographic rates, introduced recently by Hyndman and his associates, is tested on official data series of Serbian age-specific fertility rates available for period 1950-2009. The original concept of the method with its extensions and improvements is applied to region-specific data for the country (Central Serbia and Vojvodina). One of the most important benefits of the method reflected in confirmation that is essentially to model and forecast more than one principal component in order to adequately address sources of variation in fertility. Similarly, modelling and forecasting fertility rates with regards to age and not total fertility rates shows how important it is for the recognized tendency of postponing childbearing in Serbia to be included in coefficients of functional time series. Besides, the method is based completely on evaluation of historical data, without subjective views of forecasters having to be taken into account. Coherent functional product-ratio forecasts of two regions proved to be highly convergent on the long-term not allowing for outliers to contaminate the forecast.