Title: Factor modelling for functional time series
Authors: Qingsong Wang - Renmin University of China (China) [presenting]
Abstract: Functional time series now arise in many scientific fields. We consider factor modelling for functional time series. To estimate both the number of factors and the factor loading space, we develop a fully functional method by performing an eigenanalysis for a nonnegative definite matrix, formed by cross- and auto-covariance functions and the double integration. We provide both an intuitive explanation from the regression perspective and theoretical support from the asymptotic perspective. The proposed method performs well, even when the number of functional variables is relatively large compared to the number of temporally dependent functional observations. Extensive simulation studies show that our method performs well in all cases. Finally, we demonstrate the superior sample performance of the proposed methods through a real data example.