Before crunching the Predict-O-Nomics on another slate of SEC games, new TWER contributor Mac Mirable shares some tricks of the trade that should have numbers geeks — and gambling addicts — foaming at the mouse.

For those of us constantly obsessing over how well Auburn would stack up against the rest of the SEC, I’ve created a handy table – the Vegas Lines matrix – to help us obsess even more. (Click to enlarge.)
WHAT IT IS
You can use this table to get an estimate of the closing line Vegas would offer between any two teams if they played this week. For example, if Auburn were to play LSU at home this week, they would be a 2-point favorite (line of -2). If Auburn were to play Alabama on the road this week, Alabama would be a 16-point favorite (line of -16).
HOW IT WORKS
Using 70 different predictive variables, I have modeled the Vegas Line and used that model to predict the Vegas line for every possible SEC match-up. Getting a good estimate of the Vegas line is easier than predicting the actual margin of victory, but it’s still no picnic. The majority (67%) of the odds predicted so far this season have been within a touchdown of the actual odds offered in Vegas. The odds predicted above will get closer and closer to what Vegas actually offers each week, as onference games make up an increasingly large portion of each team’s year-to-date statistics. As you might imagine, it’s difficult to predict the odds Vegas will place on the Iron Bowl when Alabama and Auburn are only a third of the way through their schedules.
WHY YOU SHOULD CARE
Well, the Vegas lines are interesting in their own right and looking at how they change week after week, we can begin to get an empirical sense of how much better (or worse) Auburn is trending relative to their opponents. If that’s not enough for you, we can use these odds to get the implied probability of winning in these match-ups. These probabilities are calculated using thousands of prior games. For example, in the past 10 years (the period my data covers) there have been 212 games where the home team was favored by exactly 7 points. In these 212 games, the home team won 142, and lost 70, for an implied win probability of 67 percent.
By populating the above matrix with these probabilities, we can “simulate” the remainder of the season, and calculate such things as a team’s expected record or probability of going undefeated.
The above table (click to enlarge) reveals that Vegas likely thinks no one has a great shot at beating Alabama at home this year. Florida has an 18 percent chance of doing it, and Auburn has a fourteen percent chance of doing it. School spirit aide, if Alabama is going to lose, it’s going to have to come on the road. Of Alabama’s remaining 8 games, we’d forecast they should win 7 of them on average (with their being some small chance of their losing any of them – which adds up to their losing a total of one game). At this point in the season, they have a 22.4% chance of going undefeated.
We can also look at how Auburn’s future games should unfold. With the exception of the Iron Bowl, we should be favored in every game – except when LSU comes to town. I’m not convinced that will play out of course, but the remainder of our schedule is very favorable. Auburn should win 6 of its remaining 8 games and has a 1.9 percent chance of going undefeated. (Click to enlarge.)
Mac Mirabile is a 2002 graduate of Auburn’s economics and journalism departments. During his time at Auburn, he was a copy and photo editor with The Auburn Plainsman. He has a master’s in economics from UNC-Chapel Hill and has written numerous academic publications on college football, the NFL and gambling markets. His previous columns can be found here. Email him at macmirabile@yahoo.com.
Photo by Larry Parker.
I appreciate you taking the interest and time to compile such data. Thanks for presenting it.
So wait a minute… South Carolina, who we just beat at home, would be a 3-point favorite over us if we were to play them at home again this week? I’m afraid I don’t quite follow the Vegas logic on that one.
Yeah, Sakerlina’s entire Away column in the second table seems out of whack.
No matrix can account for how bat sh!t crazy les miles is and the butterfly effect on the college football landscape…I’m just saying.
So, by that logic, and correct me if I’m wrong, I could put $1.90 on Auburn going undefeated and win $100 if it happens? Point me to the bookie taking that bet, I’m going all in (pun intended). No seriously, I could almost double my annual salary just putting down $1000.
Those numbers are great and all, but the thing to keep in mind is that vegas lines are not a predictor of the outcome of games, they predict the public’s opinion on games, which is why they change as people bet on the games.
For the uninitiated, the idea situation for vegas is that they have an even amount of money betting on each side of the bet. Meaning, in a game where Auburn is favored 3 pts over, say, LSU, if the country has bet $30,000 on Auburn to cover, they prefer to also have people betting $30,000 on LSU to beat the spread. The reason for this is that everyone puts up a small amount of money as a sort of “insurance” for their bet. If you are to lose your bet, the sports book keeps this money.
So, if Vegas notices a trend in everyone betting on Auburn to cover, they will raise the spread to encourage people to bet more on LSU. All that vegas lines will truly tell you is how they expect people to bet.
So, seeing now that these numbers are only an indicator of how the nation expects the game to end, are you comfortable saying that this is likely how the game is going to end?
Above, you say that 67% of the time, the vegas odds are within a touchdown of your predicted odds, you already reduce confidence in your numbers by saying that its about the same as your confidence in rolling a 1-4 on a 6 sided die every time.
On top of that, you haven’t mentioned how reliably vegas actually predicts the outcome of games. You’d have to show me that the final score difference is within a touchdown of the spread at least 75% of the time before I’d even consider there to be a modicum of value in this.
Okay, so I forgot in the article to mention one crucial piece of information: If a team has a bye week or is playing a FCS opponent this week, then I am using their statistics from the prior week. That’s why SC looks like they should beat us if we replayed them. So, South Carolina, Arkansas and Miss St. are all using “old data”.
JohnnyAuburn,
I don’t know what odds a bookie would give you on Auburn going undefeated, but I bet it would pay somewhere in the neighborhood of 25/1 or 40/1. We’re 4-0, but remember we’re only a few dropped passed from being 1-3.
I probably wasn’t explicit enough in making the dilineation between what I would predict the odds of each team winning are vs. the implied odds from the Vegas spread. Certainly Vegas seeks to balance money on each side of the bet, and so these odds translate only very losely to how the season will actually play out.
I have no real response to the 67% of the time the Vegas odds are within a TD of my predicted odds — this number will only improve over time though as Vegas and my model converge on their evaluation of the ‘quality’ of each team.
Also, in the last 10 years or so, of the 7,379 games in my dataset, 2,712 (36.8%) had actual outcomes/margins of victory within 7 points of the Vegas line.
Mac, thanks for doing this. It must be nice for you to actually get to play with your data after spending what must have been (tens of?) thousands of hours over the years inputting everything.
Couple of questions: Is there something wrong with the LSU-Auburn numbers? On your point-spread chart, it looks like Auburn projects as a two-point home favorite, yet on your implied-win-prob chart Auburn is listed as <50% to win that match-up.
Also, could you speak to the home-field advantage and how it figures into your model? Vegas traditionally builds in around a three-point bonus for any home team, but in your matrix it seems to be floating in the 5- to 6-point range.
Houston Nutt,
Good catch on the on the LSU-Auburn game…same thing for SC-Florida by the way. It’s actually not an error though, more of a result of a relativelt small sample of games. These are very close games (-2 line) and as you might expect, the home team should win these games more often than not. However, in the data I have, I see the following:
-1 line: home team wins 38 of 78 times
-1.5 line: home team wins 38 of 84 times
-2 line: home team wins 31 of 63 times
-2.5 line: home team wins 61 of 147 times
-3 line: home team wins 145 of 263 times
As you can see, anything favoring the home team less than 3 points has not been a great predictor of who will actually win the game, which is why even as a 2 point home favorite, Auburn only had a 44% chance of winning the game. Now you might notice that from above 31/63 = 49%, not the 44% I used in the table. This is because I am smoothing actual win rate using data from nearby lines. So the -2 line’s 44% win rate is actually an average of lines of -1.5, -2, and -2.5. This tends to smooth out spikes in lines with a small number of observations.
Onto your second question: I think the 3 points is a good rule of thumb for most games, but I’ve tried to improve upon it in my models. I take into account stadium size, and which conference the home and away teams are from. This tends to do a better job in predicting outcomes than just using a fixed amount. Some conferences do a better job defending their home turfs than others, and my handling of the data this way allows me to understand how these various matchups effect the margin of victory. So I don’t explicitly estimate the homefield advantage effect because it’s built into several different variables I use. Interestingly, when a game is played on a neutral field, who Vegas lists as the ‘home team’ on Covers.com loses about 4 points in their predicted margin of victory.
Hope that helps, Houston.