Saturday, April 02, 2011
Sometimes it Wiggles, Sometimes it Jumps
Kevin Dowd and colleagues do some Excel and Matlab work to come up with the fact that, assuming normality, a 25-sigma event like the ones claimed to have happened in August 2007 to several of the world's largest financial companies should occur less than once in the entire history of the universe given its apparent age. So, if Kevin Viniar's claim is that a 30% decline on several days running is a >25-sigma event according to Goldman Sach's models, then the models are wrong with a certainty approaching p=1. This is the way a financial economist would look at it, anyway. An experimental psychologist or other scientist would instead say that Something Happened in August of 2007 that was not caused by random chance. Something Happens all the time in experimental science, or at least we hope it does. The distribution moves because something moved it further than can be explained by the white noise of all of the small motive causes that make everything jitter around to form a statistical distribution. Getting overly excited about which statistical distribution it is misses the point. The probability that Something really big is going to Happen again is not dependent on the shape of the distribution, and is not calculable within the distribution. It's dependent on when the conditions are right for Something to Happen again, and relovate the distribution somewhere else. That's what a "25-sigma" event tells you.