Mandelbrot Suggests A Hunt For Financial Patterns

Phoe6 writes "Wired is carrying an Open Letter of Benoît Mandelbrot, the father of the fractal, to the wizards of Wall Street, calling on them to recognize a pattern in the finantial and economic trends in the world. Mandelbrot says, If we can map the human genome, why can't we map how a man loses his livelihood? If millions can contribute a few cycles of their PCs to the search for a signal from outer space, why can't they join a coordinated search for patterns in financial markets?" I'd like to see a debate between Mandelbrot and Friedrich Hayek.

Comments

Most financial institutions already do

[EpsCylonB]

using either linear algorithmic models or a parrallel neural network approach.

However while most people agree that past performance is indicative of the future nothing can predict what is going to happen. Things such as politics and current events have a huge impact but are not easily factored in to a computer program.

There are many sophisticated solutions to recognising and predicting complex patterns but with the stock market there is an element of trying to predict the lottery.

If the lottery is run properly then every draw should be completely random, any pattern detected in past draws should be about a useful as picking your numbers out of a hat.

Re:Most financial institutions already do

[Thomas Miconi]

using either linear algorithmic models or a parrallel neural network approach.

Mandelbrot is one of the people who demolished the idea of applying linear filters (or just about any predictive method) to stock markets.

In particular he showed two things:

- Market parices are strongly non-gaussian. They do follow a bell curve (approximately as many ups than downs, with a concentratino near the middle) but if you try to calculate the mean and variance of market data and draw a gaussian curve based on it, you'll notice that the real data will have a much lower density around the mean, and much longer, higher tails at each side. This is because market price data have a distribution that favours extremes in comparison to gaussians: there are many more extreme variations (up or down) that would be expected in a gaussian process.

- Market prices have the same behaviour, regardless of the scale. If you look at a given graph of variation, it is impossible to determine wether they cover a week, a month, a year or even a decade.

Thomas Miconi

Most posters are missing the point - again.

[Schreck]

Let me recap Mandelbrot's point here.

In the April 2003 settlement of postbubble fraud charges, the biggest Wall Street firms agreed to cough up $432.5 million to fund "independent" research. Mandelbrot then makes the distinction between two kinds of research. One is the kind of research where analysts study a publicly traded company, and then give recommendations to buy, sell or hold. The other kind is fundamental economic research.

Mandelbrot then suggests that at least five percent of the settlement money be directed toward fundamental research. He does not say that we should look for a way to predict the markets with absolute certainty - that would be impossible, as many here have redundantly pointed out. (He would probably be insulted to know that so many here think that's what he advocated. He's not stupid, you know.)

He's talking about giving a boost to the kind of fundamental economic research that's already taking place. Stuff like risk management, for example. If you read the article, maybe you noticed that in the beginning he clearly gave examples of what's wrong with our present models in risk management.