B1693
Title: Bayesian inference in high-dimensional mixed frequency regression and VAR models
Authors: Kshitij Khare - University of Florida (United States) [presenting]
Abstract: Technological advancements in recent years have enabled organizations to collect, organize, store and analyze very large amounts of data from variables that are available at different temporal frequencies - e.g. monthly, weekly, daily. Such data is commonly referred to as mixed frequency time series data. First, we will focus on mixed frequency regression, where the response variable and the covariates are available at different frequencies (for example, quarterly vs. monthly). We will present a novel Bayesian methodology for (sparse) estimation of the regression coefficients and of the (autoregressive) lag length using a Bayesian nested spike-and-slab framework. Second, we will focus on mixed frequency vector autoregressive (VAR) models, which aim to capture linear temporal interdependencies among multiple time series observed at different frequencies. The issue of over-parameterization in a VAR model becomes more acute in high-dimensional settings where the number of variables is more than or comparable to the sample size. We present a Bayesian approach that achieves parameter reduction through a combination of sparsity and simple structural relationships between appropriate parameters. We will illustrate the efficacy of the proposed approach on simulated data and on real data from macroeconomics, and establish posterior consistency under high-dimensional scaling where the dimension of the VAR system grows with the sample size.
