|
Activity Number:
|
127
- SPEED: Statistical Learning and Data Science Speed Session 1, Part 1
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 29, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #301732
|
|
|
Title:
|
Comparing Time Series Graphical Lasso and Sparse VAR Algorithms
|
|
Author(s):
|
Aramayis Dallakyan* and Rakheon Kim and Mohsen Pourahmadi
|
|
Companies:
|
Texas A&M University and Texas A&M University and Texas A&M University
|
|
Keywords:
|
Multivariate Time Series;
Vector Autoregression;
Sparsity;
Graphical Lasso;
FDR
|
|
Abstract:
|
For a multivariate stationary time series, we propose sparse estimators of the inverse spectral density matrix at each Fourier frequency using joint graphical lasso. Special attention is given to the selection of the penalty (thresholding) parameters using Stein's unbiased risk estimator (SURE) criterion. The methodology is compared to the two-stage sparse vector autoregression (sVAR) modeling method in Davis et al. (2016) where partial spectral coherence measure is used to set to zeros certain entries of the autoregression coefficient matrices. We also modify the two-stage sVAR using a sparse estimator of the inverse spectral density matrix in the first stage and use false discovery rate (FDR) in the second stage. Connections between sparsity of the inverse spectral density matrix and zeros in the AR coefficient matrices are studied. We compare the two competitive methods using the simulations and real-world data.
|
Authors who are presenting talks have a * after their name.
