Parameter estimation of a mixed frequency vector autoregressive model of order 1

Abstract

For multivariate time series, different time series might be sampled at different frequencies. A mixed frequency vector autoregressive model (MF-VAR) can manage the problems arising from the mixed sampling of variables. In my thesis, I implement an estimation method for the parameters of an MF-VAR(1) model, described in Anderson et al. (2012), and investigate how the accuracy of the parameter estimation changes if the slow frequency variable is sampled less and less often compared to the fast frequency variable and if the innovation variance matrix increases. I find that the larger the distance between the observations of the slow process, the worse the estimation is.