We show that the slice sampler generates Markov chains whose variables are mean independent and thus uncorrelated when the target density is centrally symmetric. Skewness instead boosts correlations. Popular implementation algorithms such as stepping-out and multivariate-sampling-with-hyperrectangles add statistical inefficiency, the first in case of multimodality, the second in all circumstances. A new sampler which exploits these structural and algorithmic characteristics to reduce the variance of Monte Carlo estimates is experimented in several sampling problems. An insight into the properties of the product slice sampler is also provided.
PLANAS Christophe;
ROSSI Alessandro;
2024-09-13
WALTER DE GRUYTER GMBH
JRC127333
1569-3961 (online),
https://www.degruyter.com/document/doi/10.1515/mcma-2024-2012/html,
https://publications.jrc.ec.europa.eu/repository/handle/JRC127333,
10.1515/mcma-2024-2012 (online),
This document is only visible at the Commission level.
You are not authorized to publish or distribute it outside the European Commission.
This is a public document. You can share this publication.
Additional supporting files
| File name | Description | File type | |
|---|
