A Numbers Game: How Accurate are the Books?

Everybody who bets on sports is looking for that edge, a way to beat the book, a way to make the jump to a 60-plus percent success rate to beat the juice. Can it be done? Yes, but it’s getting tougher and now you’re having to compete with something smarter – technology.

Technology in Sports Betting

Long gone are the days of pen and paper (though that is what I still use) and the days of tabulating your own Excel spreadsheets. It’s a new day and age, one where you will need to create algorithms or machine learning prediction models to beat the books at a number above the “successful” rate of 58 percent. Sure, you can still hit this rate or better without models but you accomplish so much less at a much slower rate.

Let me break it down for you. First, you’ll have to find an odds data site. Once found, you will need to build a web-scraping application that can import that data into a spreadsheet. Thousands of pieces of information, imported instantly in one location for you to view. After you’ve done the data collection, you can then start playing around with creating a machine learning model to predict your own spreads and totals for games and compare those spreads against the market. Now you can find the outliers and exploit that in a much quicker way than if you were doing it by hand.

Through machine learning, you could quickly swap out any stats – total offense, rushing yards per game, points per game – virtually any and all information that could yield a response that works with analytically predicting outcomes with more precision than you could by just winging it. Or if you have enough money, you could contract a data scientist to collaborate with. Haralabos Voulgaris is an example of this. He’s known for collecting masterminds from MIT to assist in building his perfect model for finding an edge for game totals in the NBA.

These models do exist. They are owned by the Warren Sharps of the sports betting world, the elite sports betting groups and, more specifically, the sportsbooks. Ever wonder why they’ve never had a losing season? They may have had a bad (single) day but the last time a Nevada book lost over a single month was July 2013 and not once since then.

So, what’s the point of all this? All this to show you just how damn hard it is to beat the book by hand.

For context: my bread and butter has been NCAAF game totals. We’re onto Week 4 of the season, and I can’t help but think that the books have been better than in years past. Since I do write every line by hand (for record and study purposes), I’ve been noticing that totals were feeling that much harder to beat. There were many times I’d pick the OVER on a game and it would hit by the half. Now, if the OVER even hits at all, it’s on the last drive of the game and covers by a half-point. So, I wanted to see how accurate this actually was.

Is it me picking tough games to bet on or are the books improved? With the help of the Odds Shark data team (how cool is it that there’s a whole team for this?!), I reviewed 679 games from the 2018 college football season specifically referencing game totals and even more specifically, only Sportsbook lines and how they fared in the end.

After 10 hours of reviewing the data and sorting it to my liking, here’s what I found:

Game Totals from the 2018 NCAAF Season

Of the 679 games pulled from last year, 236 finished with the game total hitting within 7 points or less. That is 34.7 percent accuracy from the linemakers.  Break it down even further and of those 236 games, 110 were within 3 points of the Sportsbook line (16.2% accuracy), with the OVER/UNDER finishing at 111-118-7 in those games.

Some of these games that fell within 7 points or less included major public-betting games such as Army-Michigan (total opened at 48, finished with 45 points), Florida-Miami (total opened at 50.5, finished with 44 points) and Alabama-Georgia (total opened at 62, finished with 63 points).

Game Totals and the OVER

Of the 679 games I reviewed, 215 had Sportsbook lines of 60 or higher. This is my box. It’s these games I not only love to watch but love to bet. Life’s too short to bet the UNDER, after all. But maybe it wouldn’t hurt to take a step back and re-examine things a bit as the 2018 campaign was a volatile one for me and it was largely because the typical OVERs I would look at just weren’t hitting like they used to.

Of the 215 games with Sportsbook lines at 60 or higher, only 91 went OVER (91-120-4). To break this down even further, of the 215 games, there were 40 with Sportsbook totals of 70 or higher. Those went 18-22 on the season.

The season saw some explosive games, including: Oklahoma State over Oregon State 52-36 (opened at 72), UCF over Memphis 56-41 (opened at 72), and then there was South Carolina vs Clemson. Clemson nearly covered the Sportsbook total of 60.5 on its own, scoring 56 points. This game went way over with South Carolina scoring 35 for a total of 91.

Some other high-scoring games:

Game Totals and the UNDER

Of the 679 games I reviewed, 113 had totals of 49.5 or less. These games went 54-58-1 to the OVER. Break that down even further and there were only six games with totals of 39.5 or less. Seriously, who’s watching these games? They went 3-3. Pretty uneventful.

Some of the low-scoring affairs went way below the mark, like Arizona State over Michigan State 10-7 (opened at 44.5) and Washington over Utah 10-3 (opened at 43.5).

There were, of course, some extreme outliers that went way over their projected total:

Perhaps the most famous game to hit the OVER in 2018 was the LSU matchup against Texas A&M on November 24. Two defensive teams, how much could they have scored? One hundred and forty-six points! The Sportsbook line for this was 47 and it closed at 45.5. Instead, it went to SEVEN OVERTIMES and set multiple NCAA records, including the most combined points scored (146, in case you forgot) in an FBS football game, and it was the second-most points scored in college football history since 1937.

That record goes to Abilene Christian and West Texas A&M (NCAA Division II). That game saw 161 points scored in 2008.

What Do These Numbers Mean?

One season is not the be-all and end-all of proof but it does depict an accurate image of how the books work and that they are pretty dang good at what they do. The books should be around the 50-50 mark as they don’t make money off ticket sales, they make money from the vig. So, if they can have close to even money on both sides of the table, they’ll still profit from the juice (vig).

And there are lots of ways I could have broken this information down: by month, week, holiday weekend, by closing lines, or even by specific lines. Does 45.5 hit the UNDER more often or does the UNDER on 38 tend to go OVER?

I could have even examined the in-between totals, every Sportsbook line posted in the 50s range. Frankly, though, I got tired and that’s the point I’m trying to make. I don’t have the computing power as a machine to analyze all that data in a day. Time is a precious thing and I have spreads and totals to study, after all.

But it is still fun to break information down to see what you can find (if anything). One thing I did gather, it turns out that in 2018, life was too short to bet the UNDER OVER. Sticking to just games over 60 and under 50, the UNDER would have hit at a 54.2 percent rate – 178-145-5. That’s something, I guess.