GPU-Accelerated Bayesian Learning and Forecasting in Simultaneous Graphical Dynamic Linear Models
Department of Mathematics, Chair of Mathematical Statistics, Technical University of Munich, Munich, Germany
Technical University of Munich, 2014
@article{gruber2014gpu,
title={GPU-Accelerated Bayesian Learning and Forecasting of High-Dimensional Time Series},
author={Gruber, Lutz F and West, Mike},
year={2014}
}
We discuss modeling and GPU-based computation in a new class of multivariate dynamic models customized to learning and prediction with increasingly high-dimensional time series. This defines an approach to decoupling analysis into a parallel set of univariate time series dynamic models, while flexibly modeling cross-series relationships in a novel, induced class of time-varying graphical models of multivariate volatility matrices. A novel decoupling/recoupling computational strategy allows for exact analysis and simulation of univariate time series models that are then coherently linked to represent the full multivariate model. The computational innovations use importance sampling and variational Bayes ideas, and the overall strategy is ideally suited to GPU implementation. The analysis is enormously scalable with time series dimension, as we demonstrate in an analysis of a 400-dimensional financial time series.
September 30, 2014 by hgpu
