Discovering Graphical Granger Causality Using the Truncating Lasso Penalty

• Activity Number: 400
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Ali Shojaie*+ and George Michailidis
• Companies: University of Michigan and University of Michigan
• Keywords: Granger Causality ; Graphical Models ; Time Series ; Penalized Likelihood ; Truncating Lasso ; Large p small n

Abstract

Discovering causal relationships among random variables is the main goal in many statistical problems. However this may only be possible through randomized experiments. Time series observations provide a unique opportunity to determine how random variables affect each other over time and can be used to discover causal interactions. In this paper we propose a novel penalization method, truncating lasso, for estimation of causal relationships from time-course data based on the concept of Granger causality. This penalty can correctly determine the order of the underlying time series, and improves the performance of the lasso-type estimators. We provide an efficient algorithm for estimation of parameters and show prove variable selection consistency of the method in large p, small n settings. The performance of the proposed model is evaluated favorably in simulated, and real data examples.