A Formalized, Integrated and Visual Approach to Stress Testing

In this paper we will give for the first time a formal mathematical language to the steps used currently by financial institutions when calculating the impact of a stress scenario on a balance sheet that depends on more granular or different factors than those provided in the scenario. We will introduce the language of Probabilistic Graphical Models (PGM) – a technique rooted in machine learning – to show how the different models used at each step can be put together in a coherent picture thus giving a holistic view of the entire models setup. This will give us a solid basis to discuss some weaknesses and problems with the stress testing exercises run by the industry as of today. We will show empirical analyses to substantiate better some of our claims