Olof Heurgren

Using a Relative Standard Deviation Fixed-Width Stopping Rule as a Markov Chain Monte Carlo Diagnostic

Abstract

In this project a stopping rule ("FWSR"), designed for deciding when to terminate a running Markov Chain Monte Carlo (MCMC) simulation, is implemented and evaluated as a run length diagnostic in a piece of MCMC post-processing software called VMCMC. It is concluded that FWSR has relatively high demands in terms of run length and that it seems to favour good mixing. Discarding burn-in does not appear to have a significant effect on FWSR results.