Mardi | 2017-09-19
Salle des thèses 16h – 17h20

This paper introduces Mixed Data Sampling (MIDAS) to panel regressions suitable for analysis with GMM of the Arellano-Bond form. As MIDAS specification tests are lacking, even in univariate contexts, our proposed methods statistically embed specification checks. We proceed via confidence set estimation for the MIDAS parameter for which empty outcomes signal lack of fit. Conformably, simultaneous and partialled-out confidence sets for the regression coefficient are proposed that address the unidentified boundary parameter (Davies’ problem). Inverted statistics are asymptotically pivotal. A simulation study illustrates good size and power and, more broadly, sets a promising template for dealing with shrinkage parameters.