Title: Network analysis for time series data
Authors: Brandon Park - George Mason University (United States) [presenting]
Anand Vidyashankar - George Mason University (United States)
Tucker McElroy - Census Bureau (United States)
Abstract: Network based approaches for analyses of time series data can provide new insights concerning causality and forecasting. However, it is challenging to construct such networks using time series data, due to correlations and autocorrelations between the nodes. Additionally, the problem is getting more complicated when exogenous variables are present. We (i) describe theoretical guarantees for the regularization method for estimating the autoregressive parameters and the regression coefficients in an ARX Model, (ii) describe a new method for constructing an implicit network, and (iii) provide various network wide metrics (NWM) that are useful for identifying the active features of the implicit network. We also describe the asymptotic properties of NWM and utilize them to identify communities via the proposed implicit network.