A0450
Title: Spatial factor models for surface time series
Authors: Yasumasa Matsuda - Tohoku University (Japan) [presenting]
Runyu Dai - Tohoku University (Japan)
Yi Wu - Tohoku University (Japan)
Abstract: Surface time series is a kind of spatial panel data, i.e., panel data when a cross-sectional unit is spatially observed data. In surface time series analysis, N, the spatial dimension, is sometimes very large, while T, the time length, is usually short. It follows that popular analysis by factor models has significant difficulties in managing this feature. The focus is on surface time series as a time series of functional data on space, for which functional principal component analysis (fPCA) is introduced to define factor models. fPCA works to manage the feature of large N and short T regarding the panel as a functional time series. Spatial factor models are applied to yearly income per capita in 1700 cities in Japan for 22 years from 2000 to 2021, and it demonstrates how fPCA works to solve the problem.
