Title:Factor Models for High-Dimensional Dynamic Networks: with Application to International Trade Flow Time Series 1981-2015

Abstract:Dynamic network analysis has found an increasing interest in the literature because of the importance of different kinds of dynamic social networks, biological networks, and economic networks. Most available probability and statistical models for dynamic network data are deduced from random graph theory where the networks are characterized on the node and edge level. They are often very restrictive for applications and unscalable to high-dimensional dynamic network data which is very common nowadays. In this paper, we take a different perspective: The evolving sequence of networks are treated as a time series of network matrices. We adopt a matrix factor model where the observed surface dynamic network is assumed to be driven by a latent dynamic network with lower dimensions. The linear relationship between the surface network and the latent network is characterized by unknown but deterministic loading matrices. The latent network and the corresponding loadings are estimated via an eigenanalysis of a positive definite matrix constructed from the auto-cross-covariances of the network times series, thus capturing the dynamics presenting in the network. The proposed method is able to unveil the latent dynamic structure and achieve the objective of dimension reduction. Different from other dynamic network analytical methods that build on latent variables, our approach imposes neither any distributional assumptions on the underlying network nor any parametric forms on its covariance function. The latent network is learned directly from the data with little subjective input. We applied the proposed method to the monthly international trade flow data from 1981 to 2015. The results unveil an interesting evolution of the latent trading network and the relations between the latent entities and the countries.