A0391
Title: Forecast reconciliation and multivariate GARCH
Authors: Massimiliano Caporin - University of Padova (Italy) [presenting]
Emanuele Lopetuso - University of Padova (Italy)
Daniele Girolimetto - University of Padova (Italy)
Abstract: When considering portfolio risk forecasts, for instance, within a risk management perspective, multivariate GARCH models represent a commonly adopted approach. However, if portfolio weights are known, a univariate GARCH model on historically reconstructed portfolio returns can be considered. This represents a framework where forecast reconciliation can be considered as a tool for further improving portfolio risk prediction. By resorting to simulations, the advantage of combining univariate and multivariate portfolio risk forecasts is assessed with the aid of forecast reconciliation techniques. Results suggest relevant advantages when multivariate models are misspecified and the underlying variance is known. However, when a noisy proxy is used, all models, even the misspecified ones, are very close to each other. The analyses are complemented with an empirical example also exploiting the informative content of high-frequency data.
