Graphical models with Koehler Symanowski distributions.
Sonderforschungsbereich 386, Discussion Paper 104
In this paper, a multivariate distribution family introduced by Koehler and Symanowski (1995) is discussed as alternative assumption for graphical models which are typically connected with Conditional-Gaussian distributions. For that purpose, certain requirements which have to be fulfilled when formulating graphical models are checked. This leads to the introduction of graphical models with Koehler Symanowski distributions which are then investigated regarding some basic properties known for Gaussian graphical models.