I spent a lot of time working on the NBA predictions last weekend and I made some vast improvements to the various programs and databases I use. Now, I can make changes to the model and instantaneously see the resulting back-tested record. For example, using full-season stats, I was able to improve 5-Star pick accuracy from 57.7% to 62.4% over the time period 1/1/09 - 3/15/09. That's an improvement of 4.7 % points. Meanwhile, after back-testing 11/1/08-12/31/08, using stats up to 12/31/08, accuracy went from 66.5% (actual accuracy as the season progressed was 67.0%) to 72.1%, +5.6%.
These results are very promising, and, in the case of the '08 months, completely shocking. Granted, collective stats were used for the back-testing, but this still shows that there is potential for greater upside. These changes will be implemented immediately, including tonight's games.
On a separate note, March Madness is here! I have a similar model that fills out the bracket for me, and it has been very successful over the past 3 years. Last year it predicted the final four, finals, and champion. I'm posting the bracket tomorrow. You're free to use it as long as you're not in one of my pools. Use it as a your bracket, use it as a guide, or don't use it at all, whatever.
| TOR at CHA -7.5 |
|
| Winner: | SPLIT |
| Spread: | TOR +7.5 ***** |
| Underdog: | N/A |
|
|
| HOU at NO -3.5 |
|
| Winner: | SPLIT |
| Spread: | HOU +3.5 ***** |
| Underdog: | N/A |
|
|
| POR -6 at MEM |
|
| Winner: | SPLIT |
| Underdog: | N/A |
|
|
| SAS -5.5 at OKC |
|
| Winner: | SAS |
| Underdog: | N/A |
|
|
| NJ at DEN -13 |
|
| Winner: | DEN |
| Spread: | NJ +13 ***** |
| Underdog: | N/A |
| Correct | Games | Win% |
| STRAIGHT-UP | 1 | 2 | 50.0% |
| SU UNDERDOG WIN | 0 | 0 | ---
|
| ***** | 1 | 3 | 33.3% |
