Afterthoughts for Week 10

After debating a long time whether I should have picked Pittsburgh at home or Arizona on Monday night, I am glad I didn't. Following my lessons learned from week 9, I kept my cool, stuck to the strategy, and did not get greedy.

Although the algorithm predicted these two teams to cover the spread there were some issues. Pittsburgh had the uncertainty of Ben playing and RB Parker for sure sitting out. The Cardinals were favored, but the confidence was low (54%). San Francisco was using a new QB and Singletary was coaching them for the 2nd time. Both picks would have lost.

The strategy is to bet for games whose situation is seen previously (throughout the years) more than 20 or 30 times (big sample size) and whose confidence is at least 57%. With a confidence of 54%, I do not have enough proof to say that statistically I have found an edge. When a game is sitting at above 57% confidence of covering the point spread, then I look for injuries and other extraneous situations. If everything looks normal, as it did for the 3 picks we ended up making, then we go for the pick.

We have also learned that of the games that we should make bets for, those with higher confidence are not necessarily doing better. That is, a 58% confidence is not beating a 64% so the strategy (for now, until I figure out a better confidence measure) is to spread the wealth evenly among picks and diverse the portfolio. There are many gambling strategies in terms of the amount to put in, I will not get into that, but I will say this: do not put all the eggs in one week.


Csaba said...

Jamie, good work! I have been working on a similar model over the last 2 seasons and remarkable it has produced a 0.667 ATS success to date as well!
Have you ever considered applying the same logic/model to ther leagues?

Jaime said...

Great! Please share your weekly picks with us and/or comment on the ones I post, I'm interested in your opinion.

Yes, I have considered taking the NBA next. I think injuries affect more the results in the NBA so I may have to take a player-level approach and it would be harder to get the data. Unfortunately, time has kept me from diving into it. I also wanted to make sure that it worked well for NFL so I could apply it to NBA.

Because college has many more teams, more player turnaround, and less games, I do not think my logic would work well for college sports. I would need to rethink the approach.