A probabilistic graphical models approach to model interconnectedness

Orazio Angelini, Adrien Papaioannou, Orazio Angelini

Abstract: In this paper, we show that using multiple models when executing a specific task almost unavoidably gives rise to interaction between them, especially when their number is large. We show that this interaction can lead to biased and incomplete results if treated inappropriately (which we believe is the current standard in the financial industry). We propose the use of probabilistic graphical models – a technique widely used in machine learning and expert systems as a remedy to this problem. We discuss some numerical aspects of our approach that will be present in any practical implementation. We then examine, in detail, a practical example of using this method in a stress testing context.