CMStatistics 2021: Start Registration
View Submission - CFE
Title: In-sample inference for MIDAS regressions and GARCH-MIDAS Authors:  Onno Kleen - Erasmus University Rotterdam (Netherlands) [presenting]
Andrea Naghi - Erasmus University (Netherlands)
Michel van der Wel - Erasmus University Rotterdam (Netherlands)
Abstract: Identification issues in mixed-frequency data sampling (MIDAS) models are examined. In MIDAS models, data sampled at two different frequencies are typically linked via one single parameter. This parameter is of particular interest in determining whether there is a significant relationship between the MIDAS component and the dependent variable. However, due to nuisance parameters, the distribution of the linking parameter is nonstandard. We discuss possible ways of adjusting the test statistics in unidentified and weakly-identified cases. We apply these robust inference procedures in a simulation study of linear MIDAS models and in empirical re-examinations of previous GARCH-MIDAS applications.