Sparse Estimate of Vector Autoregressive Model

• Activity Number: 241
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Abhirup Mallik*+ and Snigdhansu Chatterjee
• Companies: and University of Minnesota
• Keywords: Sparsity ; Vector Autoregressive ; High Dimension ; Time series

Abstract

Vector auto-regressive models are generalizations of ARMA models for higher dimensions. We propose a method for finding sparse estimates of temporal dependency parameters and precision matrix in this model. This kind of model can be very useful where the component wise time series have a small number of significant inherent dependencies, for example when they form a sparse network or graph. Our proposed method can also be applied to different kinds of structures arising from dependency. We test our method using simulated data as well as try it out on real data.