Quantile Connectedness: Modelling Tail Behaviour in Financial Networks

The network images presented in this website are computed from the forecast error variance decompositions of a vector autoregressive (VAR) model estimated by quantile regression

Network statistics are obtained from the forecast error variance decompositions following the method proposed by Frank Diebold and Kamil Yilmaz (Economic Journal, 2009; Journal of Econometrics, 2014)

The model covers 18 countries and contains two credit risk measures per country -- the daily change in the five-year sovereign CDS spread and the daily change in the five-year aggregate financial sector CDS spread

The CDS spreads exhibit considerable cross-section correlation. We model this correlation using a large set of observed common factors

Having accounted for the cross-section correlation, the model can be estimated consistently on an equation-by-equation basis using standard quantile regression toolboxes. We estimate the model in R using Roger Koenker's quantreg package

Full-sample estimation results are based on a sample of 1,260 trading days spanning the period January 2nd 2006 to February 14th 2012

Where rolling regression results are reported, we use a rolling sample of 250 trading days

A rigorous technical discussion of our estimation framework can be found in the paper