The following is an excerpt from the new book The Logic of Sports Betting by Ed Miller and Matthew Davidow.
So far, the betting markets I’ve talked about are pregame markets. The main three markets — spread, moneyline, and total — and the derivative markets — ﬁrst half spread, moneyline, and total — they are all offered before the game starts. Once the game begins, the markets close.
In-play markets open once the game begins. Most sportsbooks these days offer at least some form of in-play betting. The menus at some books can be as simple as continuations of the main three pregame markets. Or at other books the menus can be extremely extensive with hundreds of betting options available at all (or at least most) times.
Say you’re watching an MLB baseball game and open your favorite sportsbook’s app. They have a couple dozen markets available on the game — moneyline, run line, total for the game, the same bets but for the ﬁrst ﬁve innings only, the ﬁrst seven innings only, alternative run lines and totals, maybe a few prop bets.
Each of these markets have prices that update in real-time as the game goes on.
If you’ve been following the theme of the book so far, one speciﬁc question may come to mind when you see these markets.
Where the heck do all these prices come from?
Live betting odds: where they come from
Most often, sportsbooks receive these odds as a data feed from a third-party vendor. There are a few established vendors who sell up-to-the-second pricing on in-play betting markets as a service. (As a matter of disclosure, at the time of this writing, I am in the process of launching one of these vendor companies. So this is a topic that I know a fair bit about, but also where I may have some personal biases that come through in the discussion.)
The way it works, roughly, is a sportsbook operator decides they would like to offer in-play betting markets to their customers. They contract with a third-party vendor who provides a data feed with markets and suggested prices. The sportsbook then adds a hold to the markets—how much they add is up to them — and they offer the bets to their customers.
Some sportsbooks contract with multiple vendors and combine the individual feeds with perhaps a twist of their own into a proprietary, aggregate feed which they then use to price the markets they offer to their customers.
Okay, so the operators buy a feed of pricing. But how do the vendor companies price all these markets in real time as the game is in progress?
An example of setting an in-game line
Exactly how any vendor does this is going to be proprietary, and they all do it a little differently. But the basic idea is that each vendor will have some sort of model, algorithm, or other process that takes in information about the teams, as well as game state information, and transforms that into prices for the markets.
Here’s a simpliﬁed example. Let’s say it’s the top of the seventh inning in an MLB game, and the home team is ahead 4‑3. The pregame betting markets closed (at market maker books) with the home team as a 62% favorite to win the game, and the total closed at 9.5 with the over at a 52% break-even percentage.
You want to make an in-play moneyline for the game. You look in your database for games that closed with similar lines (say with the home team as between a 59% and 65% favorite and totals between 8.5 and 10.5) where the home team was ahead by one run in the top of the seventh. Then you just count what percentage of those games the home team ended up winning, and there’s your line.
You wouldn’t want people to have to run these calculations during the game, so this process is automated — either by using fast database queries or by precomputing the answers.
What I just described is a simpliﬁed version of how one could make an in-play line. I don’t know how accurately this process describes what happens at the various vendors, and this sort of analysis is only one piece of the puzzle when I make a line.
But the basic idea is there. You take information about the strength of the teams (pregame closing lines), information about the game state (the score, the inning and/or how much time is remaining, and any other relevant information like who has the ball, position on the ﬁeld, and so on), and combine them intelligently to come up with a line.
Potential problems with live betting odds
There are two major problems with this, and you may already have thought of them both.
A lot of information to account for
The ﬁrst problem is that there is an enormous amount of information available once the game starts that wasn’t reﬂected in the pregame lines.
- Maybe a player or two got injured during the game.
- Maybe the teams are playing different strategies than the market expected before the game started.
- Maybe the teams are playing faster or slower.
- Maybe they’re fouling more or less.
- Maybe the weather took an unexpected turn.
- Maybe one key player is playing at a much higher (or lower) level than their usual standard.
Much of this is information that is readily available to anyone with a web browser and a TV. It’s also information that is — as you might imagine once I described how the lines get made — extremely difﬁcult (in some cases) or nearly impossible (in others) to account for correctly, in real time, second-by-second, in an in-play line feed.
And sportsbooks aren’t content these days with offering just a spread and total in-play. They want to offer dozens of markets. All updated by the second. All priced in an instant.
I call this wide range of betting options that sportsbooks offer without fully understanding them “attack surface,” a concept I’m borrowing from computer security that I think also describes sports betting well. I’ll explain the analogy a little later in the book.
The second major problem from the sportsbook’s perspective with offering in-play is that they rely on correct, up-to-the-second game state information. Since the game state is a key part of the equation to price the markets, if the game state data is either slow or error-prone, the lines priced using it will also be slow or error-prone.
In the UK where in-play betting is a more mature product, sportsbooks complain about “courtsiders,” which is a very English term for bettors who are trying to take advantage of slow data feeds by watching events in person and getting key information several seconds faster than the sportsbook receives it (via their third party odds vendor which gets it via a data feed).
Even if the data feed is fast, sportsbooks are potentially vulnerable to every glitch or mistake in the data feed. “The ball is at the Rams 35‑yard line.” Five seconds later. “Whoops. That was the Patriots 35‑yard line. Sorry about that.”
That’s ﬁve seconds the markets may have been priced for the wrong yard line — and obviously in a sport like the NFL that sort of error can make a massive difference in how markets should be priced.
Nevertheless, despite these two big problems, sportsbook operators know their customers want to bet during games, and so they offer a variety of markets priced in just this way.
How sportsbooks offer in-play bets
There’s one other big divide with how sportsbooks offer in-play betting. I’m not sure there are industry-approved terms for these variants, so I call them “timeout” betting and “running down the ﬁeld” betting.
Timeout in-play betting is a style that focuses on offering bets to customers only during stoppages in play — timeouts and commercial breaks. While the game is in progress, the bets are unavailable. Then once play stops, the markets go up for a couple minutes until the game starts back up again.
Running down the ﬁeld in-play betting is a style that always attempts to offer bets, even while the game is in progress.
Europe vs. the US
In-play betting is most popular so far in Europe, and soccer is the most popular sport to bet in Europe. Therefore, as you might imagine, running down the ﬁeld betting is very popular there. Soccer has few timeouts, so if you waited for one to offer in-play betting, you’d be waiting a very long time. And in soccer there are many periods of time during a game where the ball is being passed back and forth in the defensive half, and it’s unlikely a goal will be scored in the next thirty seconds or so. As a result, running down the ﬁeld betting is a pretty good choice for soccer betting.
American sports are different, though. Things happen faster. In the NBA, the average possession is now just a little over ten seconds. In football, a team could break a big play for a touchdown basically at any time. Even in relatively slow-paced baseball, any given pitch could turn into a game-changing home run. There’s no “downtime” in these games where nothing is likely to happen.
Not to mention hockey. Hockey is nuts.
Anyway, TV feeds are also delayed substantially. By the time you see Steph Curry bury that three on TV, likely 10‑to‑20 seconds have passed since he did it in real life, and the other team has probably already completed an entire possession in the meantime. The sportsbook knows (via their third-party vendor feed) what happened on that possession. You don’t. That’s a problem.
American sports also all happen to have plenty of timeouts. The timeout model is a great ﬁt for American sports, while the running down the ﬁeld model is fraught with problems. I’ve seen some sportsbooks try to use the running down the ﬁeld model on American sports, but to me that’s a square peg in a round hole. I suspect as the industry evolves that we will see the timeout model take root as the predominant form of in-play betting in the USA.