APBRmetrics

Crow · Joined: 20 Jan 2009 Posts: 825

Very good question.

There appear to be several issues. One of them is using the vague term "Vegas line" in the discussion.

I first referenced http://www.sportsbook.com on Sept 22 in response to back2newbelf's question on page 3 of this thread. I believe it had an opening line but I can't see what that page said then, at least immediately.

Then on page 4 Ed Kupfer cited the Painted Area article on predictions http://thepaintedarea.blogspot.com/2010/10/2010-11-nba-win-overunder-predictions.html
... and included Vegas values in his post. The Painted Area had a Vegas line of unspecified origin and date, I don't know if it was the opening line or the current line as of then. Given the Miami value of 64.5 I lean towards thinking it was the opening line of some book.

There are differences between what the PA article showed for Vegas and what Ed posted. It appears to me (now) that Ed rounded almost all of the PA stated Vegas line values down by .5 or in some cases 1 but not all for the list he posted in this thread.

I am not sure what back2newbelf used for Vegas in his tracking.

Ed **Küpfer** · Joined: 30 Dec 2004 Posts: 787 Location: Toronto

Crow · Joined: 20 Jan 2009 Posts: 825

I assumed that was the rationale and should have added that but I had something else come up to deal with.

If your listed values were used for contest error tracking, the adjustments represent a step away from that "Vegas line" but the Vegas values stated in the PA article summed to 18 wins more than available so something had to be done to address the issue by your approach or another.

If your listed values were not used for contest tracking, then your specific rounding approach is not an issue.

If the actual Vegas line with .5 win increments was used in the contest as is, then being able to pick between 2 win numbers was a possible advantage over most other predictions in the test. Most did not consider that an available or appropriate option. The difference in available choices for win predictions makes for a bit uneven average error "performance" comparison. If you want to sweat fractions or tiny fractions of a point of average error.

haralabob · Joined: 11 Apr 2007 Posts: 27

The issue is these lines are rarely 50-50 propositions.

They'll have a line like;

Mem Over 41.5 -130
under +110

this probably equates to a line of around 42.5 and not 41.5 since you are penalized by betting on the over.

Either way I'd be fairly confident that the consensus line moves by the sharps who bet these season win totals (I don't) would probably beat most of the predictors out there.

Hollingers stuff looked pretty good tho.

Jeff Fogle · Joined: 11 Jan 2011 Posts: 70

H, what do you think lies at the heart of the advantage sharps have over the various stat model "predictors?" I'd understand if you don't want to discuss something that might give away an advantage you have over the market down the road. But, if you're willing to discuss it, what elements could the models add into their mix that would improve their performance?

I know we're talking about various models and methodologies. But, at least in this thread, there are some prominent ones where the parameters are fairly well known. Why do they fall short in your opinion?

Jeff Fogle · Joined: 11 Jan 2011 Posts: 70

Oh well, thought I'd ask. Thanks for popping in with your thoughts on sharps vs. the predictors in your post.

With a little more than a week to go...

Already passed Vegas mark posted on Page 4:
Chicago, New York, Philadelphia, Indiana, San Antonio, Dallas, New Orleans, Memphis, Denver, Golden State.

Can't get there from here:
Miami, Orlando, Milwaukee, Washington, Cleveland, Portland, Utah, LA Clippers, Minnesota.

Maybe we can calculate a W-L record vs. Vegas in addition to the measurements of the misses. Those expecting big things for Golden State may have been too enthusiastic, but their predictions about the Over were correct. If we're only looking at distance away from eventual reality, we may be missing out on some indicators that did at least signal the proper direction. That might make eventual tweaks easier. Or, maybe only count differences of two wins or more just to reflect the most "opinionated" projections...

greyberger · Joined: 27 Sep 2010 Posts: 52

Find M. Haubs and put his brain in a jar. See what we can learn.

Crow · Joined: 20 Jan 2009 Posts: 825

Among the top 4 rivals to Vegas, Hollinger's predictions are closest to the rounded off version of the Vegas line from page 4 right now with an average absolute distance of 2.6, followed by me at 3.2, then new2newbelf at 3.5 and DSMok1 at 3.9.

The teams where the average difference for these 4 predictions were more than 6 wins different from this Vegas line are: Cavs, Rockets, Clippers and 76ers.

Vegas was more right so far in those cases than the top 4 rivals 75% of the time.

The edge that Vegas has on my prediction from those 4 teams is equal to 84% of the total edge it currently has on me. If you threw out those 4 teams back2newbelf and DSMok1 would move more than a half point closer to Vegas as well.

Especially if you wanted to bet, trying to build a model that mimicked the Vegas line across many years or just recent years would seem worth trying to do, in addition to trying to build your own model. It could useful even if that wasn't your purpose.

I am not sure how many 100% pure quantitative models with no manual adjustments were in the contest so that is another challenge for interpreting the strength of these basic models.

You probably want to be very confident to go more than 4-6 wins from the Vegas line. Or stop and perhaps adjust at least some.

I'd think sharps have the ability to consider whatever portion of the things I listed on page 11 in their model or after constructing the model and before betting and plenty of room to forecast changes from the past data. That might help them beat simple models or make up ground on or possibly pass Vegas in certain circumstances. A main thing for betting would be the ability to select just your best bets and not have to try to beat Vegas on all 30.