Networks, Dynamic Factors, and the Volatility Analysis of High-Dimensional Financial Series

Abstract: We consider weighted directed networks for analysing, over the period 2000-2013, the interdependencies between volatilities of a large panel of stocks belonging to the S&P100 index. In particular, we focus on the so-called {\it Long-Run Variance Decomposition Network} (LVDN), where the nodes are stocks, and the weight associated with edge $(i,j)$ represents the proportion of $h$-step-ahead forecast error variance of variable $i$ accounted for by variable $j$'s innovations. To overcome the curse of dimensionality, we decompose the panel into a component driven by few global, market-wide, factors, and an idiosyncratic one modelled by means of a sparse vector autoregression (VAR) model. Inversion of the VAR together with suitable identification restrictions, produces the estimated network, by means of which we can assess how {\it systemic} each firm is.~Our analysis demonstrates the prominent role of financial firms as sources of contagion, especially during the~2007-2008 crisis.