Did you run the G-SPAM without my apple sauce modification? Were the results any different?
I think something's off here. Factoring in volatility should increase the success rate. It's only logical. If a team plays consistently at some level, you should know how to bet their game. The application must be flawed.
Basically I'm thinking that:
1. If the game is A vs B, and
2. A plays at consistent level X, while
3. B plays at consistent level X+1,
4. if you bet B, you should expect to win.
Solution #1: Pick games in which both teams have high-ranked Beta Values (low volatility), then pick the winner based on the box's predictions. In this case we have past performance predicting future outcome, with a low possibility for variance from past performance. Overall this should yield good results.
Solution #2: Since Alpha and SHARPE ratios factor in performance and volatility in the form of a neat ratio, you should be able to pick any game and pick the team with the higher ratio as the winner. I suggest using a sample size of 10 games only, going back further than that may include irrelevant information.
*Performance = Performance vs. the average nfl team , so each team's ratios will reflect strength of schedule.
Solution #3: Creation of the Mad Capper Ratio (MCR)! When the box spits out an expected margin of victory vs. the NFL Average (MV), divide it by the Beta Ratio (BR) to come up with essentially an inflation adjusted MV, the MCR! Comparing the two teams' MCRs should give you the winner.
MCR = MV/BR