FW: JUST SPOTTED THIS TODAY

 

By chance I came across this today, I found it quite interesting and leads on to my current thinking.

Mike

 

posted by americade on Tuesday 14 Jul 2009 21:58 BST

Peter J. de Marigny, DITMo Strategies / AMERICADE (14July09)

deMarigny: Classical Statistics, The NEW Paradigm

Any student of classical statistics knows better than to discard observations in a data series that are not outliers. For instance, to a clasical statistician "Sortino" is an abhorrent measure.

However, in physics there is a law large things (like Relativity) and small things (like Quantum Mechanics). In a macro form there is meaninglessness. In a micro form there are patterns that has recently given rise to the use of "FRACTALS." So as not to turn this into a technical mathematics and physics lesson, let's just focus on the idea of a "data series" that is the central focus of risk and portfolio measurement.

Professor Nassim Taleb is the author of "The Black Swan," a book about the likeliness of outliers though represented as improbable. There is no paradigm for the prediction of these events. In a prior short article on Albourne Village I proposed that using Parametrics to predict Black Swans is like using Carbon Dating on the geological record. Non-parametrics is better applied, but there is another consideration that relates to a "data series" to predict Black Swans. I noticed that this is connected to the psychological theory called "The Hundredth Monkey Phenomenon" made popular by author Kenneth Keyes. But what is the underlying paradigm that gives rise to the idea of how trends start from nowhere, and when a Black Swan event happens?

I propose that a data series is NOT a data series at all, but that applying ideas from Quantum Mechanics (the law of small things where order is observed) we recognize characteristics hidden from the macro view of the entire data series.

I believe that in each observation there is its own data series of which it is a part. All of our statistical tools try to make sense out of the interconnection of these observations of a data series (i.e. GARCH models, etc). What is the individual observation itself is the result of its own data series so that a data series (such as a return stream from payoffs) is not an input of an independent variable in some statistical model, but is composed of the interconnection of sub-data series? That is, an observation of a data series is not part of any data series but is a data series unto itself with its own parametric and non-parametric characteristics.

If we view observations as a representation of its own data series rather than an instance in a macro data series we would have a completely different view of parametrics, Modern Portfolio Theory, and Total Quality Management that uses classical statistics as its underpinning.

Peter J. de Marigny / DITMo Strategies / AMERICADE