(Quantile) Spillover Indexes: simulation-based evidence, confidence intervals and a decomposition

Abstract
Quantile-spillover indexes have recently become popular to analyse tail interdependence. In an extensive simulation study we show that the estimation of spillover indexes is affected by a positive distortion when the parameters of the underlying fitted models are not evaluated with respect to their statistical significance, or are nor estimated subject to regularization. The distortion is reduced for increasing sample sizes, thanks to consistency of estimators, or by filtering out not significant parameters, even if in small samples it does not fully disappear due to type I error. We make another step by introducing a simulation-based approach to recovering confidence intervals from quantile spillover indexes. In addition, we put forward an algebraic decomposition of quantile spillover separating the dynamic interdependence from the contemporaneous interdependence (due to residual correlation). Empirical evidence shows that distortions on real data are sizable, and the decomposition points out that most of the spillover is due to contemporaneous effects. All of our results extend and are confirmed for the Spillover index of Diebold and Yilmaz (2009). Joint work with Giovanni Bonaccolto (University Enna Kore), and Jawad Syed Shahzad (Montpellier Business School).