A Pathological MCMC Algorithm and its Use as a Benchmark for Convergence Assessment Techniques

Abstract

We present in this paper a particular Metropolis-type algorithm for the simulation of a beta (formula) variable whose convergence is extremely slow. The interest of this phenomenon is to provide a simple benchmark against which convergence control techniques can be tested. We illustrate this use for state-of-the-art common control methods, backing up our evaluation by additional illustrations for a more standard algorithm derived from the same principle.

JEL Classification: C15