Optimal scaling of Random Walk Metropolis Hastings algorithm

Arunangshu Biswas

Abstract

In this talk I will discuss a problem of optimal scaling in the Metropolis-Hastings algorithm, which is a very useful variant of the MCMC sampler, used to simulate from any arbitrary density. I will also review the work done on the adaptive MCMC sampler or the sampler where tuning is dynamically done as the chain progresses. I will also discuss a set of properties that an adaptive MCMC sampler must have to guarantee ergodicity.