MODELS & THE RISK OF RUIN

Bill Werner is an Engineer in Missouri City, Texas. The following is his review of this year’s “worst call” and an attempt at drilling down to the ultimate problem.

~~~

>

“You will see when you can swallow the world in one gulp.”
-Zen Aphorism

>

My submittal for the worst call is that “housing markets can not fall simultaneously across the U.S. (and U.K.)” by analysts and quants that then plugged that call as an assumption into their credit default swap (CDS) models resulting in the near (!?!) collapse of the capital markets. This call based on past empirical data that regional housing markets had never gone down simultaneously before is the result of very smart people falling prey to the psychology of idiots and refusing to actually think about a problem and take the risk of ruin seriously.

This utter refusal to take the risk of ruin seriously is reminiscent of Long Term Capital Management (LTCM). If a model builder does not take into account the possibility that the model may not be perfect it is bound to fail at the worst possible time. Rather the model builder must incorporate a serious risk of ruin component in it and have the discipline to abide by it. In the final analysis there is no perfect model. And when something goes wrong in the real world, it does not care about Nobel Prizes, statistics, numerical methods, infinite-dimensional optimizations etc. All that elegance and sophistication may actually serve to blind us to the possibility of a spectacular systemic failure.

The empirical data aspect of our present error reminds me of sitting in an undergraduate macro-economics course in 1968 and the professor spouting that the overwhelming consensus from economists was that the business cycle was about to be eliminated by monetary policy. He then went on to explain that the lynchpin behind his assertion was the Phillips Curve which at the time was based on over a century of empirical data “proving” the inverse relationship between inflation and unemployment. A class of science and engineering geeks was skeptical and in today’s terminology predicted the possibility of a black swan. Policy at the time was based on the Phillips Curve and similar propositions. Ultimately the students were proven right when the inverse relationship between inflation and unemployment broke down in the 1970’s resulting in a new word to describe the phenomenon – stagflation. Paraphrasing Soros’ guru the late philosopher Carl Popper, empirical data can never rigorously prove anything.

Take a common model, a restaurant menu. A menu is a model of meals that can be ordered and eaten. Many of our model makers tend to get lost in the elegance of the models of their creation and in effect indulge in eating the menu, mistaking the model for reality. This may be harmless or result in serious indigestion or worse. A risk of ruin component has to get back to the external world in such a way that takes into account the possibility that we may be “eating the menu.” It can happen to any of us. Moreover it should be pointed out that “eating the menu” or taking a virtual model to be more real than the world we actually live in is not limited to finance but is rampant throughout business today. This may have always been the case but today’s computational power combined with lightening communications has magnified the effects to a stunning degree. This is a real “Rabbit Hole”.

If you think about it, this Rabbit Hole presented by modeling is rather obvious for finance, business and even models in science. But it is much deeper and wider than one might at first suppose. Remembering that Lewis Carroll was a master of Symbolic Logic and following the White Rabbit, it is obvious that we use the languages of mathematics to construct models, but what about our languages of daily discourse? We live in a Vast Matrix of Models of Mind Stuff that are as dangerous as they are powerful. Just look and see…