Abstract: This research proposes an ordinary least squares (OLS)-based model averaging estimator using the Mallows model averaging (MMA) criterion for the MIxed DAta Sampling (MIDAS) model. We use a Vandermonde matrix to approximate the unknown weighting functions for the MIDAS model, enabling us to semipara- metrically estimate each candidate model for averaging with the OLS estimator. We show that the proposed MMA estimator possesses the same asymptotic optimality properties considered in the literature under suitable regularity conditions, even though the data-generating process is much more general than the previously considered cross-sectional data structure. In addition to the simplicity of implementing the proposed MMA approach for the MIDAS model, our method delivers great numerical performance under various configurations considered in our Monte Carlo simulations.

Key words and phrases: Aggregate impact parameter, asymptotic optimality, model averaging, semiparametric MIDAS model.