Title: Semiparametric regularisation and estimation for partially nonlinear spatio-temporal regression models
Authors: Dawlah Alsulami - King Abdulaziz University (Saudi Arabia)
Zhenyu Jiang - University of Southampton (United Kingdom)
Zudi Lu - University of Southampton (United Kingdom) [presenting]
Abstract: Semiparametric modelling of spatio-temporal data has received increasing attention owing to its flexibility in uncovering potential nonlinear impact of a covariate on the response variable of spatio-temporal nature. For example, to model the possible nonlinear relationship between Consumer Price Index (CPI) and Housing Price Index (HPI) in United States (US), accounting for the spatio-temporal lag effects of neighbouring states would give more accurate estimation and prediction. In doing this, a fundamental difficulty is how to flexibly account for the spatio-temporal neighbouring lag effects. We propose two data-driven schemes to address solving this difficulty by extending the semiparametric regularisation methodology to simultaneously estimate and select the important spatio-temporal neighbouring lag variables for semiparametric modelling of spatial time series data. We allow the data are non-stationary over all spatial locations that are irregularly positioned on the earth surface (but stationary along time). New estimation procedures are developed with an improved family of data-driven semiparametric spatio-temporal regression models for both estimation and selection. The real data application demonstrates the proposed new models can significantly improve spatio-temporal prediction than the existing semiparametric spatio-temporal modelling.