| In econometrics and finance, variables are collected at different frequencies. If a higher frequency variable can help predict a lower frequency variable, it would be of interest to construct such regression models. One straightforward solution is to flat aggregate the higher frequency variable to match the lower frequency. However, flat aggregation may overlook useful information in the higher frequency variable. On the other hand, keeping all higher frequencies may result in overly complicated models. In literature, mixed data sampling (MIDAS) regression models have been proposed to balance between the two. In this thesis the mixed frequency models are addressed, and we propose a new model specification test that can help decide between the simple aggregation and the MIDAS model. |
