Bayesian analysis of the covariance matrix of a multivariate normal distribution with a new class of priors (Q2215742)

The paper under review deals with the problem of estimating the unknown covariance matrix of a multivariate normal population. The authors propose a new class of priors for the covariance matrix (which includes both inverse Wishart and reference priors as special cases). Their motivation is to correct the so-called ``forcing eigenvalues apart'' problem and to develop computational schemes for the new priors that allow for computational handling of large dimensional covariance matrices. Propriety and moment existence results are also obtained for both the prior and posterior distributions, and a method for subjectively eliciting the parameters of the prior is developed for a special case. Finally, the developed MCMC algorithm (for the proposed class of priors) is shown to be computationally effective for matrices of up to 100 dimensions.