An adaptive backward coupling Metropolis algorithm for truncated distributions
Markov Chain Monte Carlo, independent Metropolis algorithm, location parameter, truncated normal distribution, backwards coupling
In the paper, a new version of the independent Metropolis algorithm is considered, combining adaptive proposals with perfect sampling. construction of a non-regenerative adaptive Metropolis algorithm follows. A backwards coupling procedure is applied after each adaptation in order to guarantee the St.ationarity of the target. The result is a perfect sample from the target distribution with undesirable positive autocorrelations suppressed by adaptations. Performance of the suggested algorithm is examined using several examples of uniform and Gaussian proposals for truncated non-Gaussian targets. These examples are related to a problem of detection of the point source of a scattered signal.
Model Assisted Statistics and Applications