Practical building of subjective covariance structures for large complicated systems

摘要

Standard probabilistic Bayesian approaches frequently lead to prior representation involving systems of Gaussian unknowns, which need similar moment specifications. In practice, specifying such correlations become a daunting task to nonspecialists when no help is available. However, such help can be provided through the use of graphical models and simplified devices to make the task manageable. These ideas, put forth in this article, are based on the belief that it is easier to specify a variance than a covariance from a collection of variances. It has been illustrated with a practical example.