Sparse and time-varying covariance modeling
Hedibert Freitas Lopes, Insper
Monday, March 20, 2017 - 3:30pm
In many areas such as neuroscience, energy planning and finance there has been a growing interest in developing computationally fast methods that can estimate dependency of high-dimensional multivariate time series data. Nevertheless, the estimation of a covariance matrix based on high-dimension data is still an open problem. Another important issue is whether these covariance matrices exhibit time varying patterns. In this paper, we discuss several Bayesian regularization methods based on shrinkage and selection priors and estimate sparse covariance matrices using the modified Cholesky decomposition. Our first model considers dynamics only for the variances (stochastic volatility) and uses the Normal-Gamma prior for shrinking the regression coefficients that compose the Cholesky factor. The second model considers homoscedastic errors and time varying regression coefficients generated by a dynamic version of the Normal Mixture of Inverse Gammas (NMIG) hierarchical prior, which accommodates time varying sparsity.
Seminars generally take place in 116 Old Chem Building on Fridays from 3:30 - 4:30 pm. However, please check individual abstracts to confirm time and location. Refreshments will be served after the seminars in Old Chemistry 211. Metered Parking is available at various locations on campus. If you have never visited us before, please see our driving directions and map. The easiest and most convenient parking areas are located at the Bryan Center parking garage near Duke Statistics (recommended) or at the Sarah P. Duke Gardens. Please email or call Karen Whitesell for additional information: email@example.com or phone 919-684-8029. Sorry, but we do not have reprints available. Please feel free to contact the authors by email for follow-up information, articles, etc. Reception following seminar in 211 Old Chemistry