|
Activity Number:
|
399
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #302546 |
|
Title:
|
A Copula-Based Adaptive MCMC Sampler
|
|
Author(s):
|
Georges Tsafack*+ and Yves F. Atchade
|
|
Companies:
|
Suffolk University and The University of Michigan
|
|
Address:
|
499 VFW Parkway, Chestnut Hill, MA, 02467,
|
|
Keywords:
|
MCMC ; Simulation ; Copula ; Dependence Structure ; Multivariate Proposal ; Mixture of Distribution
|
|
Abstract:
|
In Markov Chain Monte Carlo simulation, proper scaling of the so-called proposal kernel is important for good convergence properties. There is a recent trend in the literature to design adaptive MCMC algorithms where these proposal kernels are tuned automatically to their optimal values as the simulations unfold. These algorithms hold the potential of freeing the user from the tedious parameter tuning process of MCMC simulation. But currently, these samplers are limited in that they can only capture the linear dependence between the components of the target distribution. In this paper, working in a two-dimensional space, we show that much more versatility can be achieved using proposal kernel whose dependence structure is modeled by a copula. We propose two different adaptive MCMC samplers that select the best copula to fit the target distribution.
|
