Sharp Books, Soft Books: Inside the Sportsbook Ecosystem

By Matthew Buchalter, PlusEV Analytics

Today I’m going to tackle a topic that is somewhat controversial. Myths, misconceptions and disinformation are everywhere, there is passionate debate from all directions, and there is a lot of complex stuff going on. Let’s talk about the different ways that a centralized* sportsbook can be run; specifically, how they set their odds and how they manage risk.

(* other sports betting models such as exchanges and peer-to-peer won’t be covered here.)

First let’s get some of the easier nonsense out of the way – the comparisons like “McDonalds doesn’t limit how many hamburgers you can buy”. The economics of a sportsbook are unique compared to most other industries. If you own a restaurant, all of your input costs are known at the time of the transaction, and you can set your prices accordingly – you know immediately how much you have made or lost. If you own a sportsbook and you take $110 to win $100 on the Broncos, you’ve sold something a lot more interesting than a hamburger.  You’ve sold a financial instrument, and that instrument has either a value of $210 or a value of $0 (or a value of $110 on a push). Which will it be? That is unknown at the time the instrument is sold. The sportsbook believes* that at the expected value is approximately $105, leaving the book with an expected profit of $5 – similar to what a restaurant might profit if it sells $110 worth of hamburgers. But don’t let this fool you – there is a whole world of difference between a known $5 and an unknown with an estimated expected value of $5!

(* This is a bit of an oversimplification, but the nuances of why a book would or would not set a balanced line are largely irrelevant to this discussion.)

In fact, this difference is the whole motivation for the bet to exist. The book believes the Broncos have 50% probability to cover. The bettor believes the Broncos have greater than 50% probability to cover. Each party is making their own assessment based on the information available to them. The tendency is to assume that “Vegas” is universally smarter than the public, but that’s not necessarily true. Sure they’re good at their jobs, but they have two strong forces working against them:

Optionality: A sports book has an obligation to post lines for every market, every game, every sport, every day. A bettor has an option to look at these thousands of lines and bet all of them, some of them or none of them. If a book posts 99% “good lines” and 1% “mistakes”, and if a bettor can correctly identify the 1%, then the bettor is free to pass on the 99% and hammer the 1%.

Specialization: Suppose there is a bettor out there that bets nothing but WNBA 2nd half totals. They’ve done years of research and analysis, focused exclusively on WNBA 2nd half totals. The sports book can’t afford to have someone on the payroll doing nothing but making odds for WNBA 2nd half totals, so those odds get made by a generalist who might be responsible for WNBA 2nd half totals + 99 other things. And, suppose that there are 99 other bettors out there, each of whom is focused exclusively on one of those 99 other things. You can see how the generalist can be overmatched.

The end result is a phenomenon called adverse selection that works as follows: The book believes that the Broncos have 50% probability to cover. Like all beliefs, the 50% is not an immutable fact (like the flip of a coin) but an unknown quantity that is subject to estimation error. That error could be in either direction. IF the 50% is an underestimate and the true probability is higher than 50%, and IF there are sharp bettors in the market who correctly identify it as such, then the book will take a disproportionate amount of action on the Broncos. IF the 50% is an underestimate and the true probability is lower than 50%, and IF there are sharp bettors in the market who correctly identify it as such, then the book will take a disproportionate amount of action on the Broncos’ opponents. Either way, the book’s true expected profit will fall short of the theoretical $5, and it could potentially even go negative.

Adverse selection is no joke. Insurance companies face the same types of problems, whereby they need to provide quotes for a wide range (sometimes millions!) of different risk profiles. Customers, while they don’t have the same sophisticated risk analysis capabilities, do have the optionality afforded to them by the ability to shop around and compare price quotes from multiple insurance companies – a process that is getting easier and easier for them with every new “aggregator” app or website that enters the market. Actuaries are taught at a very early stage of their education that adverse selection can pose an existential risk to an insurance operation.

So if you’re running a sports book, you need to take the threat of adverse selection extremely seriously. Over time, books have developed two very different risk management strategies as defense mechanisms:

The “sharp book” approach: Take action from everyone. Treat the flow of bets coming into your shop as valuable information, and use that information to update your lines in real time. Profile your bettors based on their past betting history, and put more weight on information coming in from sharp bettors’ plays when making adjustments. This way, if there are any errors in your lines they will self-correct as action comes in. You’ll still lose money to sharp bettors in the long run, but the amount will be controlled – kind of like plugging leaks in a boat, you’ll take on a little bit of water but you won’t sink. Sharp books and sharp bettors have a symbiotic relationship where sharp bettors are allowed to win a little bit of money in exchange for the information that they provide with their bets which the books use to sharpen their lines.

The “soft book” approach: Profile your bettors based on their past betting history. If you believe, at a defined threshold of certainty, that a given bettor is sharp, then either ban them from your book or restrict them to much smaller bet size limits than what you would offer to the general population. (Note: “Ban them” doesn’t mean “stiff them” – soft books still have an obligation to pay out any winnings that were accumulated on banned accounts, they just stop taking additional bets going forward.)

Both the sharp book model and the soft book model can be and have been run successfully. Sharp books tend to invest more in analytics and handicapping, and soft books tend to invest more in product development and marketing. There is one nice bonus that you get if you’re running a soft book…you really don’t have to put any effort into making or updating your lines at all. All you need to do is keep a close eye (or a bot) on the sharp books’ lines, copy and paste. It’s the Ludacris (but not ludicrous) approach to bookmaking – when they move, you move.

This is what happens when people say that a book is “moving on air”, i.e. moving their lines without being pushed by incoming action. They’re moving because they see sharp books moving in response to THEIR incoming action. Is this parasitic? Of course it is…but the information is available to the public and it’s not copyrightable, so it’s fair game.

Finally, at great personal risk to myself (in terms of twitter trolling) let’s explore the biggest controversy of all, the question of whether or not the soft book model is ethical.

To summarize the arguments in favour of “yes it’s ethical”: Businesses have a right (arguably an obligation in the case of publicly traded companies) to make decisions that are in their financial best interest as long as they don’t break any laws or regulations. Betting is a zero-sum game and books do not have the responsibility to accept -EV propositions for the sake of supporting advantage players. Professional bettors do not have a God-given right to earn. Denials are common practice in other industries where adverse selection is a problem – insurance companies routinely decline customers that their underwriters deem to be too high risk, and in jurisdictions where casinos have the right to refuse service they exercise that right for card counters in blackjack. This is the same argument that was made, in a very clumsy and poorly phrased way, by Barstool Sports president Dave Portnoy defending Barstool’s actions against Gambling Twitter personality Joey Knish:

To summarize the arguments in favour of “no it’s unethical”: Sports book marketing (and marketing in the gambling space in general) tends to look like this: “look at these happy winners, play at our book and you can be a winner too”. This is a bait-and-switch, because when customers deposit their money and start betting they will find that the game is rigged so that they can’t win. If they display any evidence of skill, they’re kicked out while unskilled players are given rewards and bonuses that keep them losing for as long as possible. This seems unfair and predatory, especially when viewed through the lens of the American ethos of meritocracy and freedom of commerce. I sympathize with Spanky here.

My personal opinion is that soft books are a necessary evil. To run a sharp book and be successful at it takes a high degree of skill both in data analytics and in business management, and I have the utmost respect for the companies and the people who are doing it, at places such as Circa and Pinnacle. It’s not easy. If I imagine a world without the soft book model, it’s tough to envision more than a handful of sustainable bookmaking operations left in the world. There will be heavy competition for the small supply of competent bookmaking talent, prices (in the form of vig) will be driven up and consumer choice and access will be driven down. That would be bad for everyone. Does the end justify the means? That’s a debate that I’m sure will take place on Twitter when this article is published!

My position does come with a couple of caveats. First, I find the process of a book taking a bet and “manually reviewing” it to decide whether to accept/reject it to be abhorrent. Much more unethical than banning or limiting winners. This gives the books a freeroll, they get the best of both sides by gaining the information without having to honour the bet. If the bettor is banned, they don’t attempt to place the bet and the book doesn’t get the information, it sucks for the bettor but at least there’s no asymmetry.

Second, a common argument in favour of the soft book model is “people don’t bet on sports because they are trying to win, they bet for the entertainment value / dopamine hit / whatever”. I reject this argument. I think that a sizable subset of sports bettors believes that they are skilled enough to win. It’s misguided but it’s genuine, and it shows up in other similarly complex domains. Regardless, the psychology behind the players’ choice to play does not in any way justify the game being rigged.

There is no perfect solution to this problem. The best way forward in my opinion is to have as much consumer choice and education as possible. Bet at whatever book you like. Got limited or kicked out? Find a new book and bet there, there are plenty of them out there. Or, if you’re in the “don’t get mad get even” camp, get a friend or family member to open a new account and squeeze out as much profit as you can before it gets banned, rinse and repeat.

I hope you’ve enjoyed what I’ve tried to present as a balanced, unbiased review of the sportsbook ecosystem, the square book model and the sharp book model. Feel free to contribute to the debate on Twitter!

Copyright in the contents of this blog are owned by Plus EV Sports Analytics Inc. and all related rights are reserved thereto.

2 thoughts on “Sharp Books, Soft Books: Inside the Sportsbook Ecosystem”

  1. What’s your take on the polymarket model, or the automated market maker model in general?

    Polymarket is a betting exchange but rather than traders matching the orders placed by others (a la Betfair, for example), traders interact communally with a pool of provided liquidity (typically, by polymarket or someone affiliated with them). In exchange for providing liquidity, liquidity providers earn a small fee from the traders (2% of the purchase amount, this is analogous to vig). The price a trader gets is determined by the resting price of the market, how much liquidity is being provided to the market, and how much the trader is willing to bet. These latter two points are where the system differs from a sportsbook where a price is static at a moment in time and a book will provide liquidity for bets up to their maximum for a flat price (then adjust the line after). Here, how much the bettor is betting will alter how many shares the bettor gets and how much the line moves: if placing a very large bet relative to the provided liquidity, they will move the line a lot and pay a worse price. If placing a small bet relative to liquidity, they will barely move the line at all. There is no delay between noticing sharp action and moving the line – all action moves the line automatically in proportion to its size relative to the liquidity pool.

    I can imagine such a system working for a “dumb sharp book” – when the line is first opened, the book might offer low liquidity to allow for stabilization and price discovery. After it settles, liquidity can be increased and the book earns fees as users noise trade around that line (and naturally liquidity should be increased as the closing time approaches as lines are sharper at close in general). Of course, the book can still end up at risk in any given market if the line moves hard (leaving them with the risk) and there isn’t enough fee-generating noise trading to cover the impermanent loss. (In practice, polymarket subsidizes their platform by leaving LP in *during play* which almost always means they lose as lines move dramatically in-play and they’re offering liquidity in those moments when no other sane book would).

Leave a Comment

Your email address will not be published. Required fields are marked *

We uses cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.