Non-Parametric Continuous Bayesian Belief Nets with Expert Judgement

Abstract

This report introduces continuous belief nets using the vine-copulae modelling approach. Nodes are associated with continuous distributions, influences are associated with (conditional) rank correlations and are realized by (conditional) copulae. Any copula which represents (conditional) independence as zero (conditional) correlation can be used. We present an elicitation protocol based on (conditional) rank correlations and show how a unique joint distribution preserving the conditional independence properties of the Bayesian belief net can be determined, sampled and updated.