In-Play Betting

In-play betting offers a tremendous and growing opportunity to savvy bettors. This is because it’s a hard problem for sportsbooks to solve. How do you make lines for a dozen or two games simultaneously?

Pregame lines come from price discovery at a market making book. But in-game, there’s no time for price discovery. A timeout lasts maybe two minutes. In many ways, each timeout is the same as a brand-new opening line.

These opening lines are difficult to make. It’s hard enough when the book is just trying to put up a moneyline. Say the Browns are playing at the Patriots, and three minutes into the second quarter, the Browns are up 14 to 3. But the Pats were 10.5 point favorites pregame with a 52.5 total.

Quick, what percentage of the time will the Pats come back and win? The sportsbook has two seconds to get a line up. And it needs to be within about two percent of the “right” answer or else the line will offer customers a good bet on one side or the other.

It’s not an easy problem. Fortunately for you, the bettor, it’s not your problem. No one is forcing you to bet every timeout. You can sit back and just bet the mistakes.
In general, the mistakes in these in-play lines fall into one of three categories.

1. Mistaken game state
2. Failing to account for an important game-specific factor
3. Fundamental modeling error

The first ingredient in making an in-game line is the game state. What’s the score? Which team has the ball? If the game state is wrong, the line’s also going to be wrong.

Sportsbooks get this basic information wrong more often than you might think. In football, for example, the model inputs tend to get the score wrong when something a little weird happens. A team misses an extra point. A team goes for two instead of kicking. A team kicks a field goal, but then it comes off the board due to a penalty. There’s a safety.

After all, whatever game state information is baked into the lines ultimately comes from a person watching the game somewhere and typing it in. People make mistakes. You can put error checking at points in the chain there—and for sure there is error checking—but mistakes get through.

In football they also often get the yard lines wrong. It’s common for the side of the field to get flipped, so they make the line as if the ball were at a team’s own 20 yard line, but actually they’re on their opponent’s 20. Or they get the yard line wrong because they make the line as if the result of the play counted, but instead there was a (significant) penalty. In basketball, it’s common for them to have the score wrong when they make a line.

These mistakes are all more likely to happen for lower profile games. You can probably figure out if the game you’re watching (and betting) is likely to have a modest or high error rate on the basic game state.

Over time I expect these mistakes to become less frequent. Leagues understand that their live, real-time data could be very valuable, and they also understand that to unlock the real value of the data, they have to get their error rates down to nearly zero. But, no matter what, it will likely always be worth looking for this kind of mistake.

The second mistake, failing to account for an important game-specific factor, however, will be much harder for leagues, vendors, and operators to purge from their lines. And, ironically, the more everyone wants to automate in-play betting, the more of these types of mistakes you are likely to see.

In baseball, it matters who is pitching. Duh.

Here’s an example of how an in-play data feed treated an MLB game in 2018. On July 11, 2018, the Mets hosted the Phillies. Pregame, the Mets closed at around 58 percent favorites at major market makers. Their hitting was not so great, but in this game they had the eventual Cy Young winner Jacob deGrom starting, which is why they were the favorite.

The game remained zero to zero through the end of the ninth inning, with deGrom pitching eight stellar innings. At the top of the tenth, the in-play markets were up, and the Mets were still 58 percent favorites.

Logically, this makes no sense. The Mets were only 58 percent favorites pregame because they were expected to get seven or so great innings out of the likely best pitcher in baseball. Otherwise they wouldn’t have been favorites at all.

Now he’s out of the game, the game is tied and in extra innings. Obviously, the Mets shouldn’t still be 58 percent favorites. If you happened to be watching that game and caught the in-play markets at the right time, you could have easily bagged a Phillies +130 or +135 bet that was clearly good. A bet which would have lost, by the way, as the Mets ended up winning 30 after Brandon Nimmo hit a walk-off three-run home run in the tenth.

If you know anything about baseball, this example is obvious. And the more you know about baseball, the more of these little quirks you will find. Factors like in-game weather changes, lineup changes, bullpen usage, and more that you can bet on but won’t be accounted for in the line. The same sort of things pop up in all sports. In-game injuries present opportunities like these, and it’s hard for a model to incorporate real-time injury information.

In-game strategy changes also. The NHL is a notoriously difficult sport to model in-game. There’s not much data for a model to chew on—mostly just the score and time remaining—but there’s so much more about the game that you can see with your eyes.

Leagues like the NHL are trying to bridge this gap by creating more granular, real-time player data by using wearable technologies. I think this will be something to watch going forward. But it’s one thing to have a zillion data points on where every player has been over the last five minutes. It’s another entirely to turn that into an estimated game win percentage that people can bet on. It’s far from a sure thing the models will be able to keep up with a trained human eye in hockey any time soon.

The third mistake class is more profound and harder for sportsbooks to root out of their systems. It’s basically a “the math deep inside the system is wrong” error.

Sportsbooks can compensate for the first two error types. They can do double and triple and quadruple error checking on their game state information. They can hire people who know sports and know their models and can estimate how to move lines based on injuries or pitching changes or freak weather.

But if there’s broken math in the algorithm itself, there’s not much anyone can do about that except to do a deep dive and figure out what’s wrong.

And to do that they’d have to figure out something was wrong in the first place. ♠

Ed’s newest book, The Course: Serious Hold ‘Em Strategy For Smart Players is available now at his website edmillerpoker.com. You can also find original articles and instructional videos by Ed at the training site redchippoker.com.