题目：On a Data-driven Semiparametric Time Series Model with Penalized Spatiotemporal Lag Interactions
报告人：Professor Lu Zudi（University of Southampton）
摘要：For spatial time series data, it is often an interesting but difficult task to specify the cross-sectional impacts. For example, in a study of possibly nonlinear effect of consumer price index (CPI) on the housing price index (HPI) for individual states in the United States (US), accounting for the temporal lag effects of the housing price in a given state and between neighbouring states could improve the accuracy of estimation and prediction. There lacks, however, methodology to specify objectively the spatial neighbourhoods for the identification and estimation of such spatiotemporal lag interactions. In this paper, we therefore propose a semiparametric data-driven nonlinear time series regression model that accounts for lag interactions across space and over time. A penalized estimation procedure is suggested by utilizing adaptive Lasso to estimate the important spatiotemporal lag interactions. Theoretical justification for the estimation procedure is developed. Empirical application to the US housing price data set demonstrates that the proposed method can substantially improve the estimation and prediction, while identifying nonlinear relationships between HPI and CPI and interesting spatiotemporal lag interactions.