Theory of Handicapping Odds

[SeattleJake]

Given a probability P(B) = x/y, there exists a delta such that either P(B|A) = x-d/y-d, or P(B|A) = x/y-d.

I had mild success applying this to baseball a few years back, so thought I'd share the system here.

I analyzed a set of data during a particular season (ie. not a huge a sample, but got bigger), and mostly found that besides Pitcher, there were only two consistent predictors: the last ten precentage, and the current streak. The former was a positive predictor (ie. teams tended to play the way they've been playing lately), and the latter was a negative predictor (ie. streaks tended to break). So I combined the logic of these two predictors into a single statement (ie. assuming the current streak will eventually break, the team will tend to play the way they were playing before the streak began).

P(winning|winStreak) = (last10wins - winStreak) / (10 - winStreak)

P(winning|loseStreak) = (last10wins) / (10 - loseStreak)

I compared the resulting percentages for the two teams, and bet on the team with the greater percentage. There may be a better way of applying them, but I was going for simplicity. The system consistently predicted 7-8/12 games, though when several streaks broke at the same time, tended towards 6/12. Never predicted less than 4/12, and twice returned 11/12.

Not bad. Though apparently there's something wrong with the martingale system of betting, which I wasn't aware of at the time, so I gave it up after bankrupting. I tried applying it to a small set of basketball games, and it did not have the same success, though it could have been the sample size.

Anyway, if someone gives it a try, or has the capability to run it on a statistic table, I'd be very interested in the results. Any comments welcome.