Help With My NFL Model

[dem_redbirds]

I have a question for the forum regarding a NFL model I'm currently working on. I just backtested my model on the 2008 season.

For the 2008 season vegas chalk from week 4 and out went 67.22%. My model went 63.40%. However, with my model the percentage that it gives a team winning is on the spot. Teams in the 50-59% bin win around 55% of the time, teams in the 60-69% bin win around 65%, etc.

I then took a look real quick to see how my line did vs vegas' line by looking at vegas dogs my model said should be favored. When this situation arose my plays went 23-31 su (42.6%). I also did some real quick eyeballing of games where my model gave teams a big time win% edge and the vegas line was quite small. Say last year's dallas at phila game to end the year. Phila was a 80%+ su winner according to my model. Vegas said they were small chalk. Phila crushed them. There were other games like this, as well. Then there were games were the chalk was too heavy and my model liked the dog. The dogs came in with the pts as winners.

With that said I'll now get to my question. Is this one year test enough? If not how many would be?

*As a note, the model I put together knew nothing of how the 2008 season data would turn out. All the data my model is based off is taken from previous games in past seasons.

[Justin7]

One year is nowhere near enough. 10 is a good start.

[roasthawg]

NFL is tough because of the limited number of games. The rules have changed a lot in the favor of the offense in the last decade so things that held true ten or so years ago don't necessarily work anymore. Really tough to get a feel for how a model will perform in the NFL no matter how much data you have or how you manipulate it.

[dem_redbirds]

One year is nowhere near enough. 10 is a good start.

[Wrecktangle]

Unfortunately, the yearly variation in the NFL and the lack of games makes it pretty tough. Working the NBA has gotten me to where I'm not really liking the NFL much anymore. More games, more data, more fun.

[Fishhead]

WRECK---Agree, the NBA is a money making machine.

It was really great in the 80's and early 90's.

Still, if one has a Matchbook account, any SBR poster should have no problem making money on a weekly basis with the NBA...........and many do.

[mintybetmachine]

WRECK---Agree, the NBA is a money making machine.

It was really great in the 80's and early 90's.

Still, if one has a Matchbook account, any SBR poster should have no problem making money on a weekly basis with the NBA...........and many do.

How can anyone make money at MB? You get better lines but that only gives you a few % cushion for win %.
Or did you mean just acting as the casino, thus having odds in your favor.

[dem_redbirds]

Unfortunately, the yearly variation in the NFL and the lack of games makes it pretty tough. Working the NBA has gotten me to where I'm not really liking the NFL much anymore. More games, more data, more fun.

is it still doable or should i abandon this project and move to a NBA Model? I'd like to really focus in on one sport. Then move to another.

[BigdaddyQH]

5 years is the minimum. The NBA model may help you, but the NBA is so flakey and dishonest that wagering in NBA games is nothing more than a crap shoot. For those of you who believe that the NBA had one bad official and now that he is gone, the problem is solved, I have some lovely beach front property in Kansas to sell you.

[Indecent]

is it still doable or should i abandon this project and move to a NBA Model? I'd like to really focus in on one sport. Then move to another.

Doable, yes.
Would I choose to start with the NBA instead of NFL? Yes.

[MrX]

I agree with Indecent.

[dem_redbirds]

with a nba model how many years would i have to test going forward to verify it? please, don't say 10 like the NFL.

[roasthawg]

with a nba model how many years would i have to test going forward to verify it? please, don't say 10 like the NFL.

You'll never "verify" anything... you could watch it to see if it's profitable but even then that's no assurance that it will continue to be so in the future.

[dem_redbirds]

if a dice is rolled thousands of times and it comes up "4" more than 1/6 the time wouldn't we say the dice is not fair? likewise, if you have a model that picks 54% ats when it should pick no better than 50% is it not an adequate model?

Couldn't the use of a chi-square test help in both cases in determining if what we are observing is significant or just random chance.

[Indecent]

It depends. It's quite easy to data mine and produce a model that is good at predicting games that have occurred, but terrible at generalizing for future games.

Some types of modeling are more susceptible over-fitting problems than others, but it is a reality to some extent in all forms.

[Wrecktangle]

Season to season variation is a serious problem in the NFL. As is start up weeks (1-4) wind down weeks (last two), teams in the hunt (games after Turkey day), mid-season "flatness" (October games).

Good Luck.

[dem_redbirds]

sounds like heteroskedasticity.

[Enogsiwon]

You guys ever tried watching a game?

Predicting slight changes in gameplan against upcoming personel?

[Indecent]

Predicting slight changes in gameplan against upcoming personel?

How would I get a computer to do that?

In all seriousness, I've watched less than 2 hours of sports since I started betting, and I'm hitting well above break even.

Have you tried modeling? You don't even need to know a players name on the team and you can be successful.

[Wrecktangle]

You guys ever tried watching a game?

Predicting slight changes in gameplan against upcoming personel?

I hit 60% ATS on WNBA; never watched 1 game start to finish.

Watching games is vastly over rated. After a while you start to believe the announcers and then you seriously go wrong.

[Wrecktangle]

sounds like heteroskedasticity.

pretty much, now say it quickly 5 times.

[dem_redbirds]

thank you to the man who suggested i start with the NBA. i may have just found gold. only looked at half the data for last year but through my early work i've found a 5-47 90.38% ats situation based on my numbers. forgetting about the nfl i am. this is by far the strongest situation i have ever found. i need to backtest this going way way back.

[durito]

You aren't going to beat NBA lines any easier than NFL.

Start with something with much lower limits.

[poppatom]

im interested in the way your model works.you need to give justin 7 some help.he only half answers an oppinion on who will cover.my name on the forum is poppatom.i havent posted very much.any help you supply to anyone i know will be appreciated.good luck thi8s week.