Instrumental Variables Estimation with Mixed Data Sampling (MIDAS) Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
  • Goldberger, Stephen Patterson
    • Affiliation: College of Arts and Sciences, Department of Economics
  • In most discrete time series models, the instrumental variables (IV) of estimation are the same time frequency as the error term. This is the case even if the underlying theoretical model allows instruments of a higher time frequency. This dissertation shows the asymptotic variance of IV parameter estimates may improve when instruments of a higher time frequency than the error term are utilized. In particular this dissertation shows improvements within the Mixed Data Sampling (MIDAS) framework for dealing with data series of mixed frequencies. The estimates improve in terms of asymptotic variance under three separate series of assumptions and methodologies for incorporating higher frequency instruments. First, a general methodology is outlined for constructing asymptotically better instruments for martingale difference sequence errors. Second, an argument is made for using MIDAS forecasts to construct the asymptotically optimal instruments for these martingale difference sequence errors. Finally for the the assumption of moving average errors being conditionally zero given the information set, mixed frequency information instruments lead to improvements in parameter estimates. In addition to mixed frequency instruments, considering mixed frequency moment conditions also improves estimation. In all three cases, the proposed methods are illustrated on asset pricing models using real world data.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Ghysels, Eric
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill
Graduation year
  • 2013

This work has no parents.