Bayesian Inference using Multiple-try Metropolis Hastings Scheme for the Efficiency of Estimating Gumbel Distribution Parameters

This paper aims to explore the efficiency for estimating the parameters of Gumbel simulated data using Multiple-try Metropolis algorithm (MTM). Several goodness-of-fit tests are used to compare the performance of MTM and the former, Metropolis-Hastings algorithm (MH). Concerning for a fair comparison, this study uses the equivalent starting point, the similar number of iterations and also the same length of burn-in periods. The numerical studies show that the MTM method performs slightly better than MH method after 5000 iterations to meet the stationary distribution. More candidates in the proposals lead to a higher accuracy of MTM estimation.