NBA Predictive Models

[sharpstone]

Hi everyone, I was doing some research and stumbled across this forum. I'm really impressed with the level of discussion that goes on here. I've been handicapping sports on what I'd consider to be a more serious level than most. I have a lot of other priorities in addition to capping, so learning how to build predictive models just isn't something I can take on right now. What I do is handicap the game from a fundamental standpoint and then outsource for predictive models I've found to have value. I only handicap NBA totals, so I look at a lot of pace stats to. I was wondering if anyone could please lead me in the direction of some solid public NBA predictive models or websites that provide score projections. Or if you have a model and don't post your projections, I'd be interested in talking with you.

Thank you so very much for your time and have a great Sunday!

[J.E.]

There are a couple of "team rating" sites on the internet.

Basketball-reference.com has their own SRS, which isn't great because they ignore where the games take place, and they use standard OLS regression (you can gain some accuracy by using Ridge/Lasso/ElasticNet instead)
Then there are Hollinger power rankings http://espn.go.com/nba/hollinger/powerrankings which are decent, but are probably giving too much weight to more recent games. Also, but this is one thing I probably wouldn't have brought up in any other year so I'm not sure if it's really valid criticism, it looks at a team's point differential only and ignores record. It's just weird this year because it feels like the Timberwolves are constantly ranked 8th because of their good point differential, but they again and again fail to win close games
The ones I mostly use are Sagarin's http://www.usatoday.com/sports/nba/sagarin/2014/rating/

I'll probably release one myself in the next couple of days. While I think it's going to be pretty good, I don't think one could use it to beat Vegas

[Mike G]

Sagarin was doing this before anyone else, still using the same method, and hasn't been improved upon.

[J.E.]

Sagarin was doing this before anyone else, still using the same method, and hasn't been improved upon.

He's added a new prediction/rating method called "GOLDEN_MEAN" just this season

[kohanz]

It's not public, but Haralabos ("Bob") Voulgaris uses a sophisticated prediction model that is described loosely in this article.

[boooeee]

I've got rankings on my site that are based on point spreads. Meaning that I'm trying to harness the knowledge of the betting market and reverse engineer a team ranking from that.

Mathematically, it is pretty much identical to SRS, but I'm using the Vegas point spread instead of actual scoring margin. Also, I can weight recent games much more heavily without sacrificing accuracy (although they are a bit noisy from time to time).

I haven't tested accuracy for the NBA, but the same methodology for the NFL proved to be more accurate than a broad selection of power rankings (link).

See my sig for a link to the ranking table. To get a projected margin, subtract the two team's GPF's, adjust 3.25 points for home court, and penalize the team 1.25 points if they're playing back to back.

[Barncore]

It's not public, but Haralabos ("Bob") Voulgaris uses a sophisticated prediction model that is described loosely in this article.

Ah, the things i would give to have access to his model.