We employ a battery of model evaluation tests for a broad-set of GARCH-MIDAS models and account for data snooping bias. We document that inferences based on standard tests for GM variance components can be misleading. Our data mining free results show that the gains of macro-variables in...