Modeling Network Time Series Using Generalized Network AutoRegression (GNAR)

• Activity Number: 583 - Learning Network Structure in Heterogeneous Populations
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Kathryn Leeming* and Marina Knight and Guy Nason and Matthew Nunes
• Companies: University of Warwick and University of York and Imperial College London and University of Bath
• Keywords: Time series; Network

Abstract

Network time series consist of time series recorded at nodes of a network. This network describes the relationship between the individual time series, and may feature direction or weight information. Such networks could be physical, social, or geographical, depending on the particular data application. Our GNAR model describes autoregressive behaviour of the time series both within and between nodes parsimoniously, and has been shown to outperform the VAR model in prediction tasks. The GNAR model can also capture differences in behaviour within the network, such as in different communities. A motivating environmental example will be presented, as well as possible extensions to this model.