Good points. I would like to echo two.

First, the 'soft' factors. I don't have enough past performance data so I can't provide any precision but the point is well made. Because the numbers are easy to deal with we do our analysis on those. And we've seen the numbers aren't the whole story. It's unreliable to look for high/low numbers. Patterns are often more important. And they require more work to derive. We also forget to use all of the numbers. For example, we report 30%, 70%, etc. But we don't report variance, error rate, and so on. If I told you I guarantee 50% strike with 15% POT but 700% variance the swings would destroy your bank.

Also, I have found an odd trend in the objective of computer programs. They seem to be trying to predict the win. Even the papers submitted by (apprently) super-intelligent scientists on citeseer and from universities are doing this. We aren't trying to pick the winner. We don't really care which horse wins, it's US that wants to win. If you review strategies by punters there seems to be a lot of spreading the bets around, focusing on place rather than win, tactics like dutching and so on. We really don't care who comes out ahead on the track as long as we come out ahead at the wicket (or with the bookies, as the case may be).

Well, I think that should be enough rambling for now. :smile:

-Duck

P.S. If anybody does have any historical data I would love to crunch it. Especially the soft stuff like 'rallied 3 wide' and other things that people are afraid to statistically analyze with numerical computers. If you're curious about how to analyze this stuff an easy to read book (well, easier than regular statistics books) is "Marketing Research: Methodological Foundations".