Ohio State nav bar

Seminar: Jesse Windle

Statistics Seminar Series
March 3, 2015
All Day
209 W. Eighteenth Ave. (EA), Room 170

Title

A Tractable State-Space Model for Dynamic Covariance Matrices

Speaker

Jesse Windle, Duke University

Abstract

There are several well-known tools for the Bayesian analysis of state-space models with latent states that wander around all of Euclidean space. If the latent states wander around a constrained space, like when the latent states are covariance matrices, then these tools are impaired or break down completely. I will present a state-space model that has covariance matrix-valued latent states, but remains tractable within the context of Bayesian analysis, and motivate its development and explain its utility within the context of finance.