Bayesian Robust Analysis for Large Vector Autoregression Model

• Activity Number: 143
• Type: Invited
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Hongxia Yang*+ and Ban Kawas
• Companies: IBM Research and IBM Research

Abstract

Vector autoregression (VAR) models have been widely used in economics and econometrics to analyze and predict the linear interdependencies among multiple time series. The number of parameters in VAR models is typically very large rendering the investigation of dynamic relationships computationally challenging. Even the widely used Markov chain Monte Carlo (MCMC) techniques are not able to provide a practical solution for such large-scale general model. We propose a Bayesian robust framework for large VAR models via projection techniques while accommodating uncertainty and observational noise. The resulting model accelerates model evaluations and facilitates efficient MCMC sampling in the reduced parameter space. Practical performance relative to competitors is illustrated in simulations and real data applications.