Forecasting subnational demographic data using seasonal time series methods

Forecasts of monthly demographic data are a critical input in the computation of infra-annual estimates of resident population since they determine, together with international net migration, the dynamics of both the population size and its age distribution. The empirical time series of demographic data exhibits strong evidence of the presence of seasonality patterns at both national and subnational levels. In this paper, we evaluate the short-term forecasting performance of alternative linear and non-linear time series methods (seasonal ARIMA, Holt-Winters and State Space models) to birth and death monthly forecasting at the sub-national level. Additionally, we investigate how well the models perform in terms of predicting the uncertainty of future monthly birth and death counts. We use the series of monthly birth and death data from 2000 to 2018 disaggregated by sex for the 25 Portuguese NUTS3 regions to compare the model's short-term (one-year) forecasting accuracy using a backtesting time series cross-validation approach.