Stationarity preserving and efficiency increasing probability mass transfers made possible (Q2463660)

The author proposes efficient ways of constructing transition matrices for Markov chain Monte Carlo (MCMC) samplers where efficiency has a double meaning: computational efficiency and statistical efficiency. The author uses the idea of stationarity preserving and efficiency increasing probability mass transfers (SPT) developed by \textit{A. Mira} [Stat. Sci. 16, No. 4, 340--350 (2001; Zbl 1127.60312)] to produce an efficient MCMC transition matrix, where efficiency is measured by a first-degree approximation to the asymptotic variance of the resulting MCMC estimators. A recipe of constructing the first-degree optimal matrix in proposed. An approximation of the first-degree optimal matrix is provided via an algorithm of order \(\sqrt{n}\). As an example the author show how the method can effectively improve the efficiency of a Metropolis-Hastings sampler.