WITHDRAWN: Specification Issues in Mixed-Frequency Time Series Modeling

• Author(s): Klaus Wohlrabe
• Keywords: mixed-frequency ; state-space ; model selection ; MIDAS ; forecasting

Abstract

We investigate the neglected aspect of model specification for two mixed-frequency model classes: the state-space approach and the MIDAS framework. Based on information criteria, we demonstrate that small-scale models are sufficient for mixed-frequency state-space modelling and parameter proliferation is not a major concern. In contrast, MIDAS models are very parsimonious independently of the included lags and the frequency mixture. We derive MIDAS representations from bivariate high-frequency vector autoregressions which enables us to determine theoretical lag lengths for MIDAS models. In a Monte Carlo study we demonstrate that standard selection criteria choose on average larger models than theoretically predicted. . Furthermore we outline that under certain circumstances restricted weighting functions increase forecasting accuracy.