Modeling Dynamic Transport Network with Matrix Factor Models: an Application to International Trade Flow

International trade research plays an important role to inform policy and shed light on wider economic issues. With recent advances in information technology, agencies distribute enormous amount of internationally comparable trading data, providing a gold mine for empirical analysis international trade. data can be viewed as dynamic transport network because it emphasizes the goods moving across edges. Most literature concentrates parametric modeling connectivity that focuses link formation or deformation rather than network. We take different non-parametric perspective from pervasive node-and-edge-level modeling: is modeled time series relational matrices; variants matrix factor model Wang et al. (2019) are applied provide specific interpretation Under model, observed surface assumed driven by latent with lower dimensions. Our method able unveil structure achieves goal dimension reduction. proposed dataset monthly volumes among 24 countries (and regions) 1982 2015. findings hubs, centrality, trends, patterns show matching change points policies. The also provides fertile ground future