Dynamic network signal processing using latent threshold models

Abstract

We discuss multivariate time series signal processing that exploits a recently introduced approach to dynamic sparsity modelling based on latent thresholding. This methodology induces time-varying patterns of zeros in state parameters that define both directed and undirected associations between individual time series, so generating statistical representations of the dynamic network relationships among the series. Following an overview of model contexts and Bayesian analysis for dynamic latent thresholding, we exemplify the approach in two studies: one of foreign currency exchange rate (FX) signal processing, and one in evaluating dynamics in multiple electroencephalography (EEG) signals. These studies exemplify the utility of dynamic latent threshold modelling in revealing interpretable, datadriven dynamics in patterns of network relationships in multivariate time series.

Rejoinder to ‘Dynamic dependence networks: Financial time series forecasting and portfolio decisions’

“…To Dr Scott, there is indeed interest in contextual interpretation of ordering in some applications and in using context to guide the specification, at least in part. While this was not a feature of the example in this paper, we have emphasized this point in related work with dynamic latent threshold models applied to macroeconomic time series prediction (see examples in ).…”

“…One challenge is to integrate the ‘in/out’ structures of Bayesian variable selection priors with traditional dynamic models for time evolutions, and this has proven challenging. Parallel to this theme of decouple/recouple in enabling scaling is our work with dynamic latent threshold models , where we have developed a rather comprehensive approach to precisely this problem. In a very real, practical sense, the dynamic latent thresholding concept provides for time‐adaptive variable selection , that is, dynamic sparsity: variables may come and go into/out‐of the model over time.…”

Dynamic Variable Selection with Spike-and-Slab Process Priors

“…We should like to draw particular attention to the latent threshold process of Nakajima & West (2013a,b); Zhou et al (2014); Nakajima & West (2015, 2017, a related regime switching scheme for either shrinking coefficients exactly to zero or for leaving them alone on their autoregressive path:…”

Dynamics of investor spanning trees around dot-com bubble

“…An elegant and popular method in stock market network analysis is to employ minimal or maximal spanning tree methods to find a “backbone” of the full correlation network [ 8 , 9 , 11 , 26 , 27 , 28 ]. Several more complicated correlation matrix construction and filtering methods have been developed recently [ 23 , 25 , 29 , 30 , 31 , 32 , 33 ], but utilizing these methods is left for future research.…”