On a Metropolis-Hastings importance sampling estimator
arXiv:1805.07174 · doi:10.1214/20-EJS1680
Abstract
A classical approach for approximating expectations of functions w.r.t. partially known distributions is to compute the average of function values along a trajectory of a Metropolis-Hastings (MH) Markov chain. A key part in the MH algorithm is a suitable acceptance/rejection of a proposed state, which ensures the correct stationary distribution of the resulting Markov chain. However, the rejection of proposals causes highly correlated samples. In particular, when a state is rejected it is not taken any further into account. In contrast to that we consider a MH importance sampling estimator which explicitly incorporates all proposed states generated by the MH algorithm. The estimator satisfies a strong law of large numbers as well as a central limit theorem, and, in addition to that, we provide an explicit mean squared error bound. Remarkably, the asymptotic variance of the MH importance sampling estimator does not involve any correlation term in contrast to its classical counterpart. Moreover, although the analyzed estimator uses the same amount of information as the classical MH estimator, it can outperform the latter in scenarios of moderate dimensions as indicated by numerical experiments.
arXiv comment33 pages, 4 figures, accepted for publication in Electron. J. Stat
References in corpus (9)
- Layered Adaptive Importance Sampling
- Curvature, concentration and error estimates for Markov chain Monte Carlo
- Explicit error bounds for Markov chain Monte Carlo
- Leave Pima Indians alone: binary regression as a benchmark for Bayesian computation
- Explicit error bounds for lazy reversible Markov Chain Monte Carlo
- Nonasymptotic bounds on the estimation error of MCMC algorithms
- Rigorous confidence bounds for MCMC under a geometric drift condition
- A vanilla Rao--Blackwellization of Metropolis--Hastings algorithms
- Error bounds for computing the expectation by Markov chain Monte Carlo