Steve Brooks, Uni. of Cambridge

Title: Adaptive (Reversible Jump) MCMC in Practice

Abstract:

Here we extend the work of Brooks, Giudici and Roberts who
developed a so-called efficient proposal scheme for reversible
jump MCMC. We first of all generalise their scheme so that it
is no longer restricted to transitions with deterministic "down"
moves and even show how their scheme may be implemented within the
fixed-dimensional context. We motivate these extensions as an
adaptive proposal generating scheme that tries to find the proposal
that best matches the relevant posterior conditional distribution in
the higher dimensional space. We illustrate the use of these methods for
the analysis of autoregressive time series.