This time I did a stoachastic model for the NFL.
How's the model built? It's based on 2009 data, for obivous reasons, and as the season develops I will update the model using 2010 data.
What does it simulate? It simulates the touchdowns + extra points + field goals + safeties each team will score on each game, thus obtaining the total points each team will score in each game.
How does it do it? Simple. Each team scored last year in several ways, rushing TD, passing TD, special teams TD, defensive TD, FG, Extra Points (2pt. conversion) and Safeties. Each team also allowed scores on each of these ways.
Let's look for instance at the Seahawks. They scored 7 rushing TD, 20 passing TD, 0 Special Teams TD, 3 Defensive TD, 24 Field Goals and 0 Safeties, while converting 100% of their Extra Points. So we can get gamely rates by simply dividing each of the above by 16 (total games). So the Seahawks scoring rates are the following:
Rush TD = 7 / 16 = 0.438
Pass TD = 1.25
ST TD = 0
Def TD = 0.188
FG = 1.5
Safety = 0
They also have some different rates but for their scoring allowed.
So I then combined the scoring rates of each team with the rival's allowing rates for each matchup.
For instance, the allowing Rush TD rate for the 49ers is = 0.75, so we can combine both like this: Seattle - SF Rush TD for Home Team = (0.438 + 0.75) / 2 = 0.594
So this value becomes the parameter for Rush TD for the Home team of next week's matchup between Seattle and SF.
I then simulate a Poisson distribution for each scoring type (RushTD, PassTD, STTD, DefTD, FG, Safety) using those combined parameters as the lambda of the Poisson distribution. This way, each team can only score 0 times, or once or twice, etc. according to the parameter on each scoring type.
For Extra Points the thing is different. I simluate a Binomial distribution using the total TD scored as "n" and the XP Made% as the "p".
Then I sum the total TD scored by team times 6, plus the Extra Points converted, plus the Safeties scored times 2 plus the Field Goals times 3 to get the final score of each team.
I simulated 5,000 times week 1, so I can get odds of winning for each match.
For instance, the 49ers - Seahawks game of next week has these odds:
What do you think?
Edited by Pirata Morado, 04 September 2010 - 09:48 AM.