New tools for network time series with an application to COVID-19 hospitalizations
提出了网络自相关函数和偏网络自相关函数两种新关联度量，以及Corbit可视化图，用于快速选择网络时间序列模型，并在英国NHS信托的COVID-19呼吸机床位占用数据中展示了其有效性。
Abstract Network time series models are increasingly important across many areas, involving known or inferred underlying network structure, which can be exploited to make sense of high-dimensional dynamic phenomena. We introduce two new association measures: the network and partial network autocorrelation functions and define Corbit (correlation-orbit) visualization plots. Corbit plots permit interpretation of underlying correlation structures and, crucially, aid model selection more rapidly than general tools such as information criteria. We introduce interpretations of generalized network autoregressive (GNAR) processes as generalized graphical models. We shine new light on how incorporating prior information is related to variable selection and shrinkage in the GNAR context. We illustrate the usefulness of GNAR models, network autocorrelations and Corbit plots for a novel network time series modelling of COVID-19 mechanical ventilation bed occupancies at 140 NHS Trusts. We also introduce the R-Corbit plot that shows correlations over different time periods or with respect to external covariates and plots that quantify the relevance of individual nodes. Our analysis provides insight on the COVID-19 series’ underlying dynamics, highlights two groups of geographically co-located ‘relevant’ NHS Trusts, and demonstrates excellent predictive performance.