A causal algebra for dynamic flow networks

Abstract

Over the past decades there has been considerable research to develop various dynamic forms of Bayesian Networks (BNs) and a parallel development in the field of causal BNs. However, linking these two fields is subtle. In this paper we demonstrate that, for classes of models exhibiting mass balance, it is necessary to first redefine the stochastic variables in a process using a decomposition which gives rise to a class of particular Dynamic Linear Models. These models on the transformed space can be interpreted as a dynamic form of a causal BN. A manipulation algebra is then defined to enable the prediction of effects of interventions that would not have been obtainable using the current Causal algebras. The necessary deconstruction of the processes and the algorithms to determine the effects of manipulations on the original process are demonstrated using a simple example of a supply chain in a hypothetical product market. © 2007 Springer.