A Monte Carlo Evaluation of Midas and Umidas Models for Mixed-Frequency Forecasting

Abstract

This study assesses finite sample forecasting ability of Mixed Data Sampling (MIDAS) models in the framework of a Monte Carlo simulation. Several of the restricted MIDAS specifications are compared to the Unrestricted MIDAS (UMIDAS) model in the context of a controlled mixed-frequency data generating process. Forecast accuracy and efficiency are compared for alternative sample sizes. The results indicate that restricted MIDAS models have consistently higher finite-sample performance than UMIDAS: the Beta MIDAS specification has the best finite-sample performance overall, when the true lag structure is smooth. The results offer good methodological information for the implementation of applied mixed frequency forecasting.