In the aftermath of last year's great run for compu-picks, I have decided to make a few changes to how the system will work. Aside for some formula tweaks (details are of course confidential), the main difference is in the dollar amounts being wagered.
Last year, every low confidence bet had $5 at stake (amount risked plus amount that could be won), every medium bet had $10, and every high had $15. Because it had been so successful (over a $70 return on a set of bets where $300 would have paid for the entire first two weeks' worth of bets and then some, for close to a 25% return [more if you're aggressive with how much of your bankroll you bet] ), I decided that it would be worthwhile to increase the baseline betting amounts, although of course anyone following along can scale up however much they want to.
Also, during last year the bulk of the positive returns were generated by the low confidence bets, with only a small return on the medium bets and a very small negative return (around $2.00) on the high confidence bets. It is too early to know for sure whether this was simply an anomaly or if it was a sign that the model needs to be tweaked. However, it is not too early to alter the ground rules so that this year the low confidence bets get more weight than they did before and the others less. Therefore, this time around the low confidence bets will have $10 at stake, the medium bets will have $15 at stake, and the high bets will have $20.
Furthermore, due to the success of the model picking college games last year, I have decided to create a similar model for NFL games. The ground rules for that experiment will be announced soon.