The single most important aspect of sports wagering is gauging public perception, and comparing it to reality. If you find a team that's widely overrated, there's a good chance that there will be some value in betting against them. The easy part of this is figuring out the public perception; just watch ESPN, or listen to your buddies at work talk about their picks to win it all.
The more difficult part is figuring out the real truth. For MLB, we've found that Baseball Prospectus' PECOTA is the best measure of this; in the NFL, Pro Football Prospectus is very accurate. So, with the NFL season quickly approaching, I've been thinking about quantifying both "public perception" and "reality", and seeing how that plays out on the field.
For "reality", I took the PFP win projections from the last two years, and turned those into rankings for each of the 32 teams. (This isn't perfect, since SOS is also a factor in their predicted records, but it'll have to do for now). For "public perception", I went to the ESPN.com preseason power rankings (2006, 2007). Subtracting a team's ESPN rank from their PFP rank gives us the metric we'll use; if a team is overrated by the public, their differential will be positive. I then used that as the independent variable in a regression, with the dependent variable being each team's win % against the spread for the whole season. Here's what it gives us:
You couldn't ask for much better results than these. The P-value is 0.0003, so our "Differential" variable is clearly statistically significant. And the slope is in the direction we'd expect; a positive differential means a team is overrated, so the negative slope tells us that those teams have generally performed poorly against the spread. Each rank difference is worth .004 points on win %; if a team is rated #5 by ESPN, and #15 by PFP, their difference is 10, so we'd expect them to have an ATS W% of .459.Obviously, one shouldn't blindly bet based on these numbers. But the differentials should give us an idea of which teams have value, especially at the beginning of the season. (I'd like to know if the effect is magnified in the first four games or something, but that's going to take some more data collection.) Here are the four expected ATS win percentages at each extreme for 2008:
8 comments:
FWIW, Football Outsiders hasn't really beaten Vegas the last few years.
AdvancedNFLStats.com has shown that Vegas lines (and fan expectations) don't regress enough. You can make money by betting the overs on teams predicted to be poor and betting the unders on teams expected to be really good.
If I'm remembering past Vegaswatch posts, F.O. has been about even with the books in the last few years, even after factoring in the juice? That's not a bad performance, although not earth-shattering. Also, I think F.O. is pretty new to this game, and if you read the intro to the PFP you can see they're tinkering with this any number of ways to enhance accuracy. PECOTA, on the other hand, is pretty static- which is fine by me, it does just fine.
Anyway, I think there's value in the PFP projections this year. I'm just keeping my fingers crossed that Farve goes to the Jets, so I can play the Pack and Bucs Overs.
Of course, if the Legendary Paul Bunyon.. errrr Brett Farve is traded to Tampa Bay, something tells me they will be especially public, though, regardless of the spread.
Actually, as a general rule betting the FO picks against the spread over the last three years would have gotten you smashed. I tried to track them (an inexact science since they aren't actually making picks), but I've had them batting around 45%. Part of that has been their hatred of the Cowboys biting them in the butt, but most of it is probably bad luck betting into highly accurate lines.
I don't think a comments section on this site has ever missed the point more than the one for this post.
Vegas Watch (+1)
That's some good stuff. Obviously a lot of assumptions in there, but for discussion purposes I think the model does pretty well. The results should correlate strongly with the FO o/u post you did earlier.
It sure would help if the NFL played 162 games a year. I guess we'll have to wait for
this to become reality to start getting some really good sample sizes.
I have an excel spreadsheet that tracked the 2007 regular season, except week 17, simply by their DVOA. We (my buddy who introduced me to FO and I) picked arbitrary ranges of DVOA to group them by. VW, I'll forward it if you have any interest, but I'll warn you it uses about 2% of the statistical relevance your models do.
According to my numbers, the team with the higher DVOA was .555 ATS, while the team with a higher DVOA as an underdog were .609 ATS, and road dogs .634 ATS.
Post a Comment