The Mixed Data Sampling variance estimator is a novel procedure able to overperform the univariate GARCH models and other conventional methods in empirical applications. We propose a suitable estimator for the multivariate context which is easily computable and tractable even in large-scale problems. We address the one-step-ahed forecasting accuracy at the monthly frequency over alternative models. These include the unconditional sample estimator, the rolling-window (or realized) estimator, and a constrained multivariate GARCH model. The MIDAS multivariate procedure provides a signi.cant reduction in the out-of- sample bias over these alternatives.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados