Skip to main content

Graphical Modelling in Multivariate Time Series and Matrix Data Analysis

I will discuss some of our recent work in dynamic modelling for multivariate time series and sparsity modelling in matrix data analysis. This includes dynamic matrix-variate graphical modelling for portfolio investment management in financial time series, with examples drawn from individual/personal (i.e., "you and me")-level mutual fund studies as well as the more typical institutional/investment bank-oriented exchange rate studies. Central topics include questions of graphical model uncertainty and search, and the decision-level implications of sparse model structuring. Similar concerns arise in problems of modelling matrix data - i.e., contexts in which each single observation is a matrix - and I will discuss some of our recent work with novel graphical models and some examples in macro-economic time series as well as novel Markov random field models of potential interest in spatial and imaging problems.

Collaborators include Duke StatSci graduate student Hao Wang, undergraduate student Craig Reeson, and alum Carlos Carvalho (University of Chicago).