Title: Network modeling of high-dimensional time series in the presence of factors
Authors: Sumanta Basu - Cornell University (United States) [presenting]
Abstract: Estimating network connectivity among multiple time series is an important problem in many economic and financial applications. Examples include macroeconomic forecasting, system-wide risk monitoring and portfolio optimization. Sparsity-inducing regularizers commonly used in high-dimensional statistics do not account for the presence of pervasive factors influencing the underlying dynamics. We address the problem of estimating the networks of intertemporal and contemporaneous connectivity among multiple time series in the presence of common factor processes. For models with observable factor processes, we propose a regularized maximum likelihood procedure to estimate the factor loadings and conditional independence structure among the idiosyncratic components. In the presence of latent factors, we propose a low-rank and sparse modeling strategy to estimate the network after accounting for the underlying common factors. We demonstrate the advantage of the proposed methods on numerical experiments and a motivating application from financial economics.