Abstract:
|
There has been recent progress in the estimation of high-dimensional sparse Vector Autoregressive (VAR) models. However, in certain applications, selective time series are collected at different frequencies, thus giving rise to the need to model such mixed frequency data. This talk considers a Gaussian VAR model for such data, wherein the high frequency time series are modeled as multiple time series, so as to much the low frequency sampling of the remaining ones; for example, in the presence of monthly and quarterly data, each monthly time series is modeled as three time series, thus leading to an expansion of the parameter space. Sparsity is introduced for the model parameters through a spike-and-Gaussian slab prior distribution. Under certain regularity conditions, we establish consistency for the posterior distribution under high-dimensional scaling. Numerical experiments on synthetic data illustrate the performance of the estimates obtained from the mean of the posterior distribution and the coverage of their respective credible intervals.
|