A0671
Title: A dynamic network autoregressive model for time-varying network-link data
Authors: Jingnan Zhang - University of Science and Technology of China (China) [presenting]
Abstract: Network-linked data, where different units are linked through a network, has been extensively studied in the literature. However, its extension, specifically time-varying network-link data, has received less investigation. Existing methods for time-varying network-link data only assume that units' attributes change over time, neglecting network evolution. To address this gap, a dynamic network autoregressive model is proposed for time-varying network-link data, where both units' attributes and networks are allowed to vary over time. A tensor decomposition method is employed to provide low-dimensional embedding vectors, which are further used to reformulate the traditional network autoregressive model. Interestingly, node-embedding vectors are concentrated around some group centers but are not exactly the same within some groups. Meanwhile, both within-group and global homogeneities are considered for the effect of covariate vectors. To tackle the resultant optimization task, the power update algorithm and an efficient alternative updating algorithm are developed. Furthermore, the asymptotic consistencies of the proposed method are established, irrespective of the presence of the global effect of the covariate vector. These consistencies are demonstrated by extensive simulated examples and a real example of time-varying network-linked fund data.
