:: Workshop – Bayesianism II: Foundations and Applications ::

Invited Talk 7 (thursday – 15h00)
New Perspectives On Bayesian Dynamic Models: Foundations And Applications.
Fabio Rigat (The University of Warwick).

Abstract:
Unlike for other fields, Bayesan modeling of dynamics processes is currently an area where parametric assumptions provide the most popular avenue to model building and model selection. Standard parametric dynamics, either linear of non-linear, are often specified using conjugate priors allowing changes in the parameters'values at each time point. With the advent of high–frequency data in finance and biology, this approach may lead to inferring fast and irregular dynamics which are difficult to interpret in the application context. This limitation calls for generalisations of the methodology admitting tick–by–tick dynamics as a limiting case. In this talk we review recent approaches to defining Bayesian dynamic models in semi-parametric and non-parametric frameworks. These involve the use of infinite-dimensional prior processes, dynamic factor models and change-point models based on appropriate test statistics. We then proceed to illustrate the application of the latter approach to modeling dynamic neural processes and multivariate daily financial time series, contrasting inferences and predictions with those of their fully parametric limiting cases.

Promotion and Support:
			CAPES-Proex Pós–graduação em Estatística-USP