re: NFL betting models: useful or mostly BS?

quote:

---

I have 1 that comes up with a point spread using penalties, line, turnovers and point differential. Most of the time it's close. If I see a 4pt difference I look into that game. If it's spot on I might not touch it. It works some years 60% and others it will be around 50%.

I bet mostly underdogs. I really would like to have a rating on every player so I could use injuries but I don't know of a website that rates Olineman and Dlineman, LB and secondary

---

That’s basically the kind of direction I’m going with mine too. I don’t think a model is ever going to be perfect, but if it can find where the line looks off and also tell you when not to touch a game, that’s valuable. The injury part is tough because everyone knows QB matters, but OL, DL, LB and secondary injuries can swing a game too and most people don’t really price that in right. I’m trying to build mine where player impact actually affects the projected spread, not just shows up as a note.

quote:

---

The book already does that for you. That's how they get their number. You think you know more than them?

---

Books are sharp, no doubt. But if their first number was always perfect, lines wouldn’t move and closing line value wouldn’t be a thing. I’m not trying to beat every number. I’m trying to find the few they may have hung a little light and pass on the rest.

How are you testing your model to tell if it will be effective vs just noise?

quote:

---

I’m not trying to beat every number. I’m trying to find the few they may have hung a little light and pass on the rest.

---

Go heavy on CFB bets you love in the first few weeks of the season.

Vegas also seems to struggle a little with college baseball spreads if you really wanted to devote time to a model

I have heard of this new fangled thing called AI.

I think the bigger hurdle in making a profit in today’s sports betting industry is the inflated vig. It didn’t used to be this bad. I feel like the golden age of online sports betting was the early 2000s. I really miss the 8 to 10 cent lines on baseball they had back then. I’ve stayed away since then because I can’t stomach the 20-25+ cent lines. Ugh.

Alright Brendon, quit bragging!

Models fail when injuries hit late on Sunday morning. You need clean line data before overcomplicating injury algorithms.

Tracked a bunch of lines last season when late scratches were destroying my plays, and checking odds on Fanatics Markets helped me spot where the sharp money was actually flowing before placing anything. Definitely publish picks before kickoff, backtests lie easily.

To you original question, yes people will pay for picks in the NFL, I know this because I run my own website with subscribers. The problem with NFL and NBA for generating models is the model is already made up of public information the books already know and have priced in, the key is finding niche edges within that information. And the most important part that you will find when following X picks is most people have no reliable backtest or leaked backtest (AI is notorious for this) and haven't run there actual edge against a randomized/null control test. When you think you have found an edge with your NFL have AI run this test on your backtest data, if it beats the random pull test then you have something. I have written several articles about this but can't post my site here as will get whacked/anchored.