Afterthoughts NFL Week 15 - Point Spread Predictions
Another bad week for picks not so bad for point spread predictions. Let me explain. When I first started making building a model that could predict point spreads I, I wanted it to be as precise as possible given team-level data. That is, using team records, game results, home field advantage, and offensive/defensive stats I build a bunch of models in order to pick the most accurate. That was step 1.
The next step was to determine which of these predictions was best to use in making picks. The obvious candidate was those predictions that were most different from the point spread. This theory was thrown out because predictions were way off when there were key injuries or teams had clinched playoff spots at the end of the season.
I knew it would be a waste of time to try to pick every game. So the goal was to find a way to use the predictions to make picks. That is when I built the now useless confidence levels you will not see posted on the blog again. These confidence levels were merely the percentage of times predictions picked correctly against the spread in the past 7 years given the point spread and the difference between the point spread and the prediction.
I have noticed the past 3 weeks that the original idea works well if one manually disregards games with key injuries (something I cannot do automatically given the data I have). Lets take a look at this past week point spread predictions posted on the previous blog. These are the 6 games with the biggest difference in point spread vs the Vegas spread. Disregarding the Giants/Dallas game due to injuries to Burress and Jacobs all picks won except the Washington/Cincinnati game and Seattle/St. Louis was a push. Unfortunately, it is hard for me to go back to historical data and test this theory, but until I have a better confidence measure, this is what I will use to make picks from now on. Last year, I made a huge comeback in the playoffs, only making one incorrect pick. There's still hope.
The next step was to determine which of these predictions was best to use in making picks. The obvious candidate was those predictions that were most different from the point spread. This theory was thrown out because predictions were way off when there were key injuries or teams had clinched playoff spots at the end of the season.
I knew it would be a waste of time to try to pick every game. So the goal was to find a way to use the predictions to make picks. That is when I built the now useless confidence levels you will not see posted on the blog again. These confidence levels were merely the percentage of times predictions picked correctly against the spread in the past 7 years given the point spread and the difference between the point spread and the prediction.
I have noticed the past 3 weeks that the original idea works well if one manually disregards games with key injuries (something I cannot do automatically given the data I have). Lets take a look at this past week point spread predictions posted on the previous blog. These are the 6 games with the biggest difference in point spread vs the Vegas spread. Disregarding the Giants/Dallas game due to injuries to Burress and Jacobs all picks won except the Washington/Cincinnati game and Seattle/St. Louis was a push. Unfortunately, it is hard for me to go back to historical data and test this theory, but until I have a better confidence measure, this is what I will use to make picks from now on. Last year, I made a huge comeback in the playoffs, only making one incorrect pick. There's still hope.
| Game | Vegas Line | Estimate | Difference |
|---|---|---|---|
| PITTSBURGH @ BALTIMORE | -2 | 2.9 | 4.9 |
| WASHINGTON @ CINCINNATI | 7 | 12.6 | 5.6 |
| SAN FRANCISCO @ MIAMI | -6.5 | 2.9 | 8.4 |
| NY GIANTS @ DALLAS | -3 | 5.7 | injury |
| SEATTLE @ ST LOUIS | 3 | 8.6 | 5.6 |
| BUFFALO @ NY JETS | -9 | -4.3 | 4.7 |
| NEW ENGLAND @ OAKLAND | 7 | 11.3 | 4.3 |
Comments