How would your portfolio fare if the Euro were to break up next month? Would the degree of diversification that you assumed in its construction still hold? Would the realized returns (say, of Bunds, BTPs or OATs) bear a passable resemblance to what you assumed? Would even the sign of the assumed correlation remain valid?

Nobody really knows, of course. But what we do know is that it would be of little use looking for answers in past stress events, as crises always end up unfolding in their unique idiosyncratic way, and circumventing the Maginot lines built by looking at what happened during the last crisis.

To make sense of the opening question, a number of ancillary questions are required. Why did the Euro break up? Was it an orderly unwinding or a panic-lead demise? Was it occasioned by some other concurrent crisis? As the question is contextualized, the narrative immediately becomes richer and more complex. If one wants a believable and actionable scenario analysis, this complexity is unavoidable, but it comes at a high cognitive and computational cost.

This is where a number of techniques offer valuable assistance. The Bayesian-Net Technology is particularly promising in this respect, because it is mathematically and logically solid, but extremely intuitive, and because it lends itself to sensitivity analysis and to critical interrogation by an intelligent but not-mathematically-versed decision maker. The idea underpinning the approach is to connect via conditional probability tables a root event (a GDP shock in the super-simple example in the figure below) to a number of transmission channels, and, ultimately, to portfolio sensitivities in a logically consistent and cognitively resonant way.


What ‘drives’ the Bayesian-net construction is the choice of the causal links between the root causes and the portfolio sensitivities and the specification of the attending conditional probabilities. These probabilities can be sourced from a variety of sources, such as expert knowledge, market-implied estimates or statistical data. It is exactly the richness and variety of these inputs that allows the user to handle possibly unprecedented scenarios, and to create correlations and dependencies potentially very different from the historical ones.

There has recently been a lot of work in translating these promising ideas into practical applications[1] and much, of course, remains to be done. However, the field is so exciting, and already mature enough, that EDHEC-Risk Institute is launching a public-domain site dedicated to investment solutions, to be released in September/October 2017, where users are guided to build their own Bayesian Nets to stress the portfolios of their concern. The Roman satirist Juvenal made famous a certain metaphor for an impossible event: rara avis in terris nigroque simillima cycno — as rare a bird on earth as a black swan. After black swans were discovered to be abundant in Australia more than a thousand years later, they have become a spirit animal for the worst events in the financial world. They are seen as unpredictable, disastrous, and almost mystical. Bayesian Nets cannot turn black swans into white geese – but, at least, can help us making us better prepared when the next flock of birds, of whichever colour, appears in the sky.

The Bayesian-Net Technology is particularly promising in this respect, because it is mathematically and logically solid