Research on GARCH-type multivariate volatility and correlation models typically concentrates on accurate estimation and prediction of return dynamics. To a much lesser extent, the literature has focused on the economic forces at play in the background and potential volatility transmission mechanisms. We propose to extend the classical MGARCH specification for volatility modelling by developing a news tone based structural MGARCH model targeting structural identification of volatility spillovers. Instead of relying on widely used ad-hoc decomposition techniques of the conditional covariance matrix based on ordering of variables, we suggest, similar to the proxy-SVAR framework, to obtain the structural decomposition of the conditional covariance matrix of a system of asset returns using instrumental variables. In the spirit of sentiment econometrics, which has sparked considerable progress in estimation and forecasting of financial volatility in recent years, we investigate news tone based measures as proxy variables for identification of our structural MGARCH model. In an empirical application, we assess the the impact of structural shocks by investigating volatility impulse response functions based on economic turmoil events.
