Last year I calculated the implied team ratings for every FBS team given the Vegas over/under win totals. That generated good discussion, so I figured I would do it against this year. What I did boils down to the following idea: if Vegas thinks each team is going to win n games (based on their adjusted over/under*), what does the rating (i.e. strength) of each team need to be to such that based on their schedule their expected wins is n? In other words, you build a model that links team strength to probability of victory in a game, such that when you sum all the game probabilities you minimize the error between modeled team wins and Over/Under projected wins.
(*) the Over/Under values I used were the Vegas Insider numbers posted here and have been adjusted up or down to reflect the money line over/under odds included. The schedule was pulled today from sports-reference here. If anyone wants to get into an esoteric technical discussion on how this was done or how the entire model works I'm happy to do that.
I am not claiming that the ratings produced are perfect reflections of what Vegas thinks. In fact, these win totals are from the end of May (I couldn't find any more recent values, only national championship odds) so I know this isn't what Vegas currently thinks (or more properly the betting market thinks). Baylor has an O/U of 9.5 games, but as well all know it is a mattress fire in Waco right now. So if anyone has a source for more recent O/U totals for each team, I'd be happy to update the model.
Anyway, not a ton of surprises here. Vols come in at #7, though their rating is almost equivalent to #8 Michigan.
Another cool thing you can do with this is get the implied spreads of each game. Again, I don't think these are perfect by any means - we know the Vols are 23 point favorites week 1 against App State but this system puts it at 19.5 - but I think they are interesting to look at nonetheless.
Note: Win% represents the Vols probability of winning the game. The spread for each game is listed in the normal way, i.e. negative values represent home favorites, positive values home dogs. The VaTech game is obviously neutral site, but Vols are an 8 point favorite according to this.
For a comparison check out the spreads that the system produced last year. Overall it did pretty well IMHO.
Alright, I'm done. Hopefully it generates some good discussion and not a bunch of comments about how nerdy I am. I already know that.
(*) the Over/Under values I used were the Vegas Insider numbers posted here and have been adjusted up or down to reflect the money line over/under odds included. The schedule was pulled today from sports-reference here. If anyone wants to get into an esoteric technical discussion on how this was done or how the entire model works I'm happy to do that.
I am not claiming that the ratings produced are perfect reflections of what Vegas thinks. In fact, these win totals are from the end of May (I couldn't find any more recent values, only national championship odds) so I know this isn't what Vegas currently thinks (or more properly the betting market thinks). Baylor has an O/U of 9.5 games, but as well all know it is a mattress fire in Waco right now. So if anyone has a source for more recent O/U totals for each team, I'd be happy to update the model.
Anyway, not a ton of surprises here. Vols come in at #7, though their rating is almost equivalent to #8 Michigan.
Another cool thing you can do with this is get the implied spreads of each game. Again, I don't think these are perfect by any means - we know the Vols are 23 point favorites week 1 against App State but this system puts it at 19.5 - but I think they are interesting to look at nonetheless.
Note: Win% represents the Vols probability of winning the game. The spread for each game is listed in the normal way, i.e. negative values represent home favorites, positive values home dogs. The VaTech game is obviously neutral site, but Vols are an 8 point favorite according to this.
For a comparison check out the spreads that the system produced last year. Overall it did pretty well IMHO.
Alright, I'm done. Hopefully it generates some good discussion and not a bunch of comments about how nerdy I am. I already know that.
