Daily Speculations

The Web Site of Victor Niederhoffer & Laurel Kenner

Dedicated to the scientific method, free markets, deflating ballyhoo, creating value, and laughter;  a forum for us to use our meager abilities to make the world of specinvestments a better place.



Write to us at: (address is not clickable)

Phil McDonnell

Review of a Geologist's Selective Predictions

The French geology professor was predicting another U.S. Stock market slam down in May of 2003. If you listened to him you were short in one of the biggest bull markets of the last century. Despite having read his book I still ignored his advice and somehow was fortunate to eke out gains higher than anyone has any right to expect for the year.

In my opinion the flaws in the professor's approach are multiple. He looks at tails of distributions, the extraordinary events which occur rarely and analyzes only those - a suspect procedure.

He changes his model to fit every different market. When he fits the Hong Kong market he uses one model, the Nikkei is fudged with a different model and the US Markets are fitted with another. One wonders why the Holy Grail is constantly changing.

The work uses a common and interesting thread of criticality. Specifically this means that as a market approaches a critical bubble peak the risk and the reward of that market go to infinity - an unsustainable event. This is a very interesting idea borrowed from chaos theory. However fitting terms involving 1/(t-c) where c is the critical date is extremely problematic. As the time approaches c the numbers approach infinity and the variance approaches infinity squared. Regressions and other fitting models which minimize squared error become infinitely unstable. Sornier demonstrates this in one section of the book when he tries to predict c and its value oscillates wildly with each new fit - sometimes overshooting and sometimes undershooting.

Data fits are performed on data which has not been detrended - a mistake which no self respecting spec would make. Generally goodness of fit / ANOVA data is not provided, although in defense I would comment that any such calculation would be suspect because of the multicollinearity, non-linearity of the underlying functions and possible infinite variance associated with the criticality term.

The bottom line for me is that underlying method is suspect but that the professor has interesting ideas in the areas of criticality, crowd following and contagion. Based on subsequent out of sample results the predictions have been, well, catastrophically wrong. Even the latest Nikkei-US Market comparison is suspect as I note that each time the Nikkei bottomed it coincided with a top in the US Market and each time the Nikkei topped it coincided with a bottom in the US. This happened both before and after the top in 2000. In this case even the in sample predictions are disastrous.