Something interesting to think about: the old models are about 80% correct when picking the winner straight up. The new model is about 60% ATS. On previous nights it has been the case that when a) the old model picks the winner, and b) the new model picks the opposite team to win, given its point predictions. The old model's straight up picks have been correct while the ATS pick has been incorrect.
The model is one week young. This is a learning process, maybe we can find some trends that will improve the accuracy. Tonight, that's the situation with the SAS-UTA game. Let's see what happens!
| POR -4.5 at NYK, 203 | |
| POR | 105.6 |
| NYK | 97.9 |
| Winner: | SPLIT |
| Spread: | POR -4.5 |
| O/U: | OVER |
| GS at OKC -7, 222 | |
| GS | 108.1 |
| OKC | 112.7 |
| Winner: | OKC ***** |
| Spread: | GS +7 |
| O/U: | UNDER |
| SAS at UTA -1.5, 192.5 | |
| SAS | 102.8 |
| UTA | 101.5 |
| Winner: | UTA |
| Spread: | SAS +1.5 |
| O/U: | OVER |
| 7-Dec | Correct | Games | Win% |
| SU - non ***** | 1 | 1 | 100.0% |
| ATS - non ***** | 0 | 3 | 0.0% |
| O/U - non ***** | 2 | 3 | 66.7% |
| SU ***** | 1 | 1 | 100.0% |
| ATS ***** | 0 | 0 | --- |
| O/U ***** | 0 | 0 | --- |
| SU Overall | 2 | 2 | 100.0% |
| ATS Overall | 0 | 3 | 0.0% |
| O/U Overall | 2 | 3 | 66.7% |
8 comments:
no plays in the nuggets sixers game?
The line wasn't up when I posted- something about injuries and Iverson. I'll re-evaluate whether or not it's worth posting in a couple hours.
How much/long did you back test the system for?
860 games over the 2008-9 season. This new system is dynamic; it continually adapts to changes in home court advantage, amongst other variables.Changes in HCA absolutely destroy any static model.
Very Interesting. I wish you luck with this in the future. I will definitely pay attention.
Do you think 860 over only 1 season is enough to give an accurate representation of the model's accuracy? My experience(though limited) with statistics, models and sports shows that a system may work for 1 or 2 years but be shown to be faulty over a 7 or 8 year period. What do you think? Are you currently back testing it for other seasons?
I can test the 07-08 season without too much difficulty, I just need a weekend to work on it.
Collecting the necessary data for prior seasons is too labor intensive, so right now 07-08 is as far back as I plan on going.
Looking forward to hearing what the results are for the 07-08 season. I'm working on a few things myself. Here's to hard work hopefully paying off!
Post a Comment