A stochastic multi-regional model for Italian population projections

Francesco C. Billari, University of Oxford
Gianni Corsetti, Istituto Nazionale di Statistica (ISTAT)
Rebecca Graziani, Università Bocconi
Marco Marsili, Istituto Nazionale di Statistica (ISTAT)
Eugenio Melilli, Università Bocconi

In this work we show the results that emerge from the first attempt to produce expert-based stochastic regional forecast of the Italian population for the period 2013-2065. We apply the method proposed by Billari et al. (2012), where the full probability distribution of forecasts is specified on the basis of expert opinions on future developments of the main components of the demographic change. In particular, we derive the joint forecast distribution of the pair Total Fertility Rate and Immigration, and of the pair Male and Female Life Expectancies at. The forecast distributions of Emigration and Mean Age at Birth are derived separately. The conditional elicitation procedure makes it possible to elicit from experts information on the future marginal behaviour of a single indicator in terms of expected value, variability and correlations. We designed a questionnaire according to such elicitation procedure and submitted it online to thirty Italian demographers (Billari et al., 2013) on the future national trend. The aim of this work is to asses this stochastic methodology in the framework of the so-called “multiregional model”. The forecast of future regional trends of the above mentioned demographic indicators is derived in a framework of convergence scenario of the Italian regions towards the mean expert-based forecasts. It is also assumed that the relative variability and the correlations between indicators are constant across regions and equal to those derived at the national level from application of the elicitation procedure. Twenty-one regional forecast probabilistic distributions are so obtained for each summary indicator. Also internal migration are forecasted following a probabilistic approach by using a multiregional matrix. Finally 2,000 samples are drawn out from the corresponding multivariate distributions, so to obtain, in a simulation based approach, the forecasts of the population sizes (by age and sex) for the twenty-one Italian regions.