Title: A model-based spillover index
Authors: Massimiliano Caporin - University of Padova (Italy) [presenting]
Giuseppe Storti - University of Salerno (Italy)
Abstract: A generalized version of the spillover index is derived from a conditional autoregressive Wishart model (WAR). The WAR model is first estimated by Quasi Maximum Likelihood under L1-penalization and taking advantage of the analytical gradient. The spillover index is then computed, accounting for the interdependence between realized variances and covariances. To recover the index, a novel rectangular forecast error variance decomposition is introduced, assuming shocks on N stock returns and a reaction on the $N$ realized variances and the $0.5N(N-1)$ realized covariances. Empirical examples contrast our index, with and without interdependencies in the realized covariance modeling, to the original parameterization based on a Vector Auto-Regressive (VAR) model including only realized variances.