Title: Tensor Decomposition for Financial Network Analysis

Abstract: Nowadays, networks represent a widely analysed topic in financial research.  Networks’ topology is explored and measures of connectivness and centrality are extracted for each stock (node). The estimated network relies on the sample used to build it. Hence, by construction, it is conditioned on time horizon used and the sampling scheme will affect the results at any level of the network analysis, possibly leading to different conclusions. In this paper we propose a method which relies on tensor decomposition allowing to separate the structural part of the financial network, cleaned by time specific events, from its dynamic one. The first object can be used to assess systemic risk while the second one can be used in evaluating idiosyncratic (dynamic) risk.