Speaker: Matteo Barigozzi, University of Bologna

Title: FNETS: Factor-Adjusted Network Estimation and Forecasting for High-Dimensional Time Series

Abstract
We propose FNETS, a methodology for network estimation and forecasting of high-dimensional time series exhibiting strong serial- and
cross-sectional correlations. We operate under a factor-adjusted vector autoregressive (VAR) model which, after accounting for pervasive co-movements of the variables by common factors, models the remaining idiosyncratic dynamic dependence between the variables as a sparse VAR process. Network estimation of FNETS consists of three steps: (i) factor-adjustment via dynamic principal component analysis, (ii) estimation of the latent VAR process via l1-regularised Yule-Walker estimator, and (iii) estimation of partial correlation and long-run partial correlation matrices. In doing so, we learn three networks under- pinning the VAR process, namely a directed network representing the Granger causal linkages between the variables, an undirected one embedding their contemporane- ous relationships and finally, an undirected network that summarises both lead-lag and contemporaneous linkages. In addition, FNETS provides a suite of methods for forecasting the factor-driven and the idiosyncratic VAR processes. Under general con- ditions permitting tails heavier than the Gaussian one, we derive uniform consistency rates for the estimators in both network estimation and forecasting, which hold as the dimension of the panel and the sample size diverge. Simulation studies and real data application confirm the good performance of FNETS.

Area of research: Econometrics

Host: Leopoldo Catania

Organisers: Luke Taylor and Leopoldo Catania.