A0616
Title: Homogeneity pursuit in forecasting high-dimensional functional time series: Is clustering necessary
Authors: Chen Tang - The Australian National University (Australia) [presenting]
Han Lin Shang - Macquarie University (Australia)
Yanrong Yang - The Australian National University (Australia)
Yang Yang - Monash University (Australia)
Abstract: Joint modelling and forecasting high-dimensional functional time series (HDFTS) has begun to gain popularity in the literature. However, heterogeneity would deteriorate prediction accuracy. In this vein, pursuing homogeneity within the HDFTS is the key to improving forecast accuracy. Two methods are compared in the effort to pursue homogeneity: one is through the clustering framework based on a functional panel data model with fixed effects, and the other one is through a dual-factor model for HDFTS. Different from the functional panel data model with fixed effects, where the homogeneous part and the heterogenous part are additive, the dual-factor model puts the collection of functional time series in a lower dimension (homogeneous) and a group of population-specific basis functions (heterogeneous) as a product, which avoids clustering. An empirical study shows that the proposed model produces relatively accurate point and interval forecasts for age-specific mortality rates in 32 countries. The financial benefits associated with the improved mortality forecasts are translated into a life annuity pricing scheme.
