Tennis quant betting

by on January 18, 2014 at 7:25 am in Games, Sports | Permalink

For someone who says he bets millions of dollars on tennis a year, sports gambler Elihu Feustel doesn’t watch many matches.

“Which one is Granollers?” Feustel says, referring to Marcel Granollers, a Spaniard ranked 35th in the world. “Is he the one that’s good on clay courts?”

Feustel, from South Bend, Indiana, says he doesn’t need to pay attention to who the players on the men’s ATP World Tour are to double his money. He relies on an algorithm he created using data from 260,000 matches to make about 30 bets a day on Grand Slams such as the Australian Open, which started Jan. 13.

Gamblers and investment funds are increasingly vying for profits from tennis by using computer models to win money from more casual bettors, according to Scott Ferguson, a former Betfair Group Plc (BET) education officer. Such quantitative analysts, or so-called quants, are focusing on tennis in the same way their counterparts are employed by hedge funds to predict moves for stocks, bonds and other assets.

Betfair, a London-based company that enables bettors to wager against each other online, matched almost 50 million pounds ($82 million) of bets on the 2012 final in which Novak Djokovic beat Rafael Nadal. Djokovic is an 8-11 favorite to win a fourth straight title in Melbourne with U.K. bookmaker William Hill Plc, meaning a successful $11 wager would return $8 plus the original stake.

Granollers prefers clay courts, according to his men’s tour profile, and lost his first-round match with Marin Cilic of Croatia in five sets on the second day of play on the hard courts of this year’s Australian Open.

…Tennis is an “attractive” sport to create an algorithm for because there are only two players in a singles match and statistics are freely available, according to William Knottenbelt, an associate professor of computing at London’s Imperial College. He co-wrote a tennis algorithm that he says would have made a 3.8 percent return on bets on 2,173 ATP matches in 2011.

Feustel, who says he puts in a 60-hour week checking and improving his model, works with a computer programmer and trader. The programmer trawls the Internet for data such as serve speed and break-point conversions. That’s plugged into the model which comes up with “fair” betting prices for scheduled games.

If those odds diverge from market prices, Feustel says, his trader — who lives outside the U.S. — will gamble as much as the market will allow at bookmakers including Pinnacle Sports, based on the Caribbean island of Curacao. That can be about $30,000 on a match result in later tournament rounds.

There is more here, and for the pointer I thank Hugo Lindgren, who is joining Hollywood Reporter as acting editor.

Z January 18, 2014 at 9:23 am

I’ve been tinkering with a model for picking football games. Over the last two years I’ve been 57% against the spread on college games. That’s cherry picking the games that jump out. Expanding the selection lowers the winning percentage. NFL games are much more difficult as the volume of betting is higher. The syndicates that have been using quants to beat the book make their money finding the opportunities to bet both sides of the game. There’s few games, but better odds.

Zachary Mayer January 18, 2014 at 10:02 am

Out of curiosity, what kind of model are you using?

There’s an interesting Kaggle competition out there right now for sports betting: http://www.kaggle.com/c/march-machine-learning-madness

Z January 18, 2014 at 2:32 pm

In all candor, “model” is an overly generation term in my case. Sports gambling is the oldest form of crowd sourcing. Winners pay losers so you’re not betting against the house. You’re looking for where the crowd is mistaken. Professional sports gamblers often make a lot of money on famous teams like Notre Dame or the Dallas Cowboys. The reason is the stupid money follows these teams, resulting in a line that is way out of range of reality. The Notre Dame – Alabama game a couple of years ago is a good example. It went off at 9.5.

Something not mentioned is the book makers are wise to all of this stuff. They have quants of their own who analyze betting patterns to try and find the robots. Unlike the stock market, the casinos don’t want the machines making bets. There’s a book called The Smart Money written by a guy who was a runner for a betting syndicate. Picking winners was not as hard as eluding the counter measures by the casinos.

Matt Rekoske January 24, 2014 at 10:36 pm

Your post is full of contradictions and misinformation. If you don’t have a model, how are you identifying the games where the crowd is mistaken? And if you’re not betting against the house, then why would the casinos even care about quants or robots making bets? Hint: some books care, some don’t.

Also The Smart Money dealt with late-1990s Las Vegas, which is completely irrelevant to the current offshore betting landscape.

Squarely Rooted January 18, 2014 at 10:09 am

This is a very interesting case because it highlights the difference between pure gambling and even the most reckless speculation on financial products – the latter affects the outcomes of what is being speculated on. So regardless of what you think about “hedging,” “speculation,” etc, the markets in for example oil futures are driven by an underlying commodity and supply and demand for that commodity; speculation on oil futures change the price of oil which change usage patterns and real patterns of resource utilization and distribution.

Gambling on sports, however, doesn’t affect the outcome of the underlying event (or at least shouldn’t, though of course it does sometimes, but let’s ignore that for now). Which means that massive inflows into say tennis gambling can’t be balanced by changing the odds offered by bookies (which are fixed) or by affecting who wins the match the way that massive inflows into say the British Pound can result in this: http://en.wikipedia.org/wiki/Black_Wednesday.

This means that gambling on sports is purely redistributive, without creating any kind of additional surplus or efficiencies the way other financial markets are at least purported to; it also means that these kind of algorithm-driven inflows are unsustainable, since they require counterparties. Either bookies will go under or progressively adopt odds to the predictions of the algorithms (assuming they are superior to existing methods). But in the meantime no other political or economic outcome has been effected; it was simply a temporary transfer from bookies to algorithm-based gamblers. It didn’t result in an adjustment of prices to better reflect underlying economic realities that then resulted in more efficient utilization and distribution of resources. It’s just gobbling the dollars of foolish, or at least insufficiently prepared, bookies.

Scoop January 18, 2014 at 10:34 am

Could you explain that further please?

As I understand it, bookies don’t make money by betting on sports. They make money by matching up people who want to make bets by shifting the odds/spread until the number of dollars bet on one side matches the other.

I’m sure that they are often forced to take positions because they can’t make sides match exactly or because the shifting odds/spreads leave holes. But that’s not the plan.

Or am I just wrong?

Ken Rhodes January 18, 2014 at 11:29 am

You are correct, Scoop. The way they “balance the two sides” depends on accurately setting the early line so that the inflow of bets is approximately equal on the two sides. Thus, the importance of the oddsmaker, whose job it is, not to predict the outcome of the event, but to predict the inflow of bets as a function of the spread, so that the early line doesn’t have to change much, if at all.

Consequently, the best opportunities for the quant bettor arise when the early line diverges significantly from the predictions of the quant model. Loading up on a spread that does not accurately represent the likelihood estimate for the event gives one immediate opportunity, and occasionally offers a second opportunity–a straddle that occurs when the early line has to change significantly to keep the total inflow balanced.

Steven January 18, 2014 at 12:11 pm

I remember reading a model on this a while back (~15 years ago).

Fixed pool of biased bettors.
Unlimited pool of unbiased, profit-maximizing bettors.
Unbiased, profit-maximizing bookies.
Fixed vig/commission.

Bookies did not set the spread/odds to balance the two sides. Rather the set the spread/odds in such as way that they ended up taking a position against the biased bettors. The extent to which they could exploit the biased bettors was limited by the existence of the unbiased bettors and the vig/commision. They set the spread/odds for maximum exploitation, subject to the constraint that the edge given to the unbiased bettors had to be less than or equal to the house edge created by the vig/commission. The author(s) applied the model to some betting market (professional football?), and found some evidence that this was actually how the spread/odds were selected.

I may be misremembering some of the details, but that should be enough for anyone who cares to track down the paper (with a bit of google-fu).

Charlie January 18, 2014 at 5:15 pm

Maybe this is what you are referring to?

How Do Markets Function? An Empirical Analysis of Gambling on the National Football League
Steven D. Levitt

http://www.nber.org/papers/w9422

Z January 18, 2014 at 2:39 pm

The defect in your reasoning is the bookmakers are not wagering on the games. They provide the market for others to wager. Losers pay the winners and they pay the bookmaker. That’s called the vig. You make a bet on team X and put down $110. Someone else bets the other side for $110. The winner gets his $110 back, plus $100 from the loser. The Bookie gets the $10 from the loser.

That said, you are correct that betting on the Lakers does not change the Lakers or their game. Betting on Amazon does change the behavior of Amazon as well as others who may or may not be betting for or against Amazon. That’s because we pretend the bet on Amazon is an investment.

Squarely Rooted January 18, 2014 at 3:39 pm

So here’s the basics plight of the bookie, as far as I can see it:

If I want to go long on a financial position, I have to find someone else to go short, and vice-versa. Large exchanges and liquid markets mean that at any given price if I want to take a position, long or short, I will probably find a counterparty.

Sports betting markets aren’t like that, even though they want to be like that. Instead, the bookie is everyone’s counterparty – essentially, the bookie is as short as everyone who wants to be long and as long as everyone who wants to be short. This puts the bookie in both an enviable and risky position – enviable because he could receive a windfall if the “market” is wrong, risky because he could lose his shirt if the “market” is right. However, the bookie is also taking a commission. This is the key to his solution – he is both counterparty and intermediary.

Therefore, the bookie wants the market price to be one where longs and shorts roughly balance out, so that he is up all his commissions and has no other losses. Rather than think about “odds” the easiest way to think about this is with football “spreads” in which the bookie does his best to decide how much to adjust the “real” score to get bets to cancel on the “modified” score.

In the end, the bookie is essentially relying on his expertise, both at understanding the sport as well as understanding gamblers.

Now what happens if you unleash mega-powerful computers that can basically calculate more accurate outcome distributions than bookies? You will have a flood of bets onto one side. This lopsidedness is exactly what bookies don’t want, because if the lopsided side of a bet prevails it is the bookie who eats the losses.

Ergo, these algorithms are essentially creating bookie-to-algorithm creator transfers, transfers that will cease, presumably, once bookies start using the algorithms themselves.

Doug January 18, 2014 at 3:56 pm

What you’re describing is implicit in any market. Bookies operate as liquidity providers, in financial markets we call them “market makers.” Market makers must always take some inventory risk, even if they perfectly price everything they will at some points end up net long or short because they need to satisfy immediate demand. If they’re mostly interacting with “dumb money” its no big deal. Their positions will mostly be random and over the long-term have no correlation to the outcome. Hence the return they earn by selling at higher prices than they buy compensates them.

In contrast if they interact with “smart money” the situation gets bad. The smart money has a net edge of which side is more profitable, hence the market makers wind up with “adverse selection.” In other words the inventory they build up tends to move against them. Liquidity providers everywhere and always try to segment their customers to seek out the dumb money and avoid the smart money. Bookies might do this by cutting off betters that are too good. Stock market making algos might do this by flashing their orders at random times or pulling their quotes during market moving events.

Ultimately though the transfer isn’t from bookie-to-algorithm. If the proportion of “smart money” rises to “dumb money” bookies will lose more money initially. However in equilibrium eventually the spreads between what they buy and sell at will increase to compensate them for this cost (to the extent that they can’t reproduce the algorithms themselves or segment the informed bettors out). So the cost is ultimately borne by the uninformed regular bettors.

BC January 18, 2014 at 11:55 pm

You are correct in identifying the bookies as the market makers. However, I think that “smart money” like Z can actually help them by allowing them to offload inventory risk, which they don’t really want anyways.

The profits for both “smart money” and bookies comes from what you call “dumb money”, but which we might less judgmentally call price-insensitive or line-insensitive bettors. These bettors place bets primarily for entertainment value rather than as part of a profit seeking business. As Z says, they will bet on Notre Dame, Dallas, or whoever out of emotional attachment rather than as the result of a business process. Bookies take the opposite side of these bets and try to adjust their lines (prices) to balance the bets on both sides so that they can earn their vig while minimizing inventory risk. However, when entertainment demand is too strong on one side, then the bookies’ inventory gets too high until they have adjusted the lines enough to attract “smart money” like Z. At that point, the point where the “line that is way out of range of reality” to use Z’s terms, the bookies are able to offload some of their inventory risk to “smart money” like Z.

The result is that the line-insensitive bettors are able to bet on their favorite teams at better lines than would have been the case in smart money’s absence. (Without the smart money, the bookies would have to keep adjusting the lines to keep their inventory risk from getting too big or to get compensated for bearing the extra inventory risk.) So, their cost of entertainment becomes a little cheaper (in terms of their expected bet value). The bookies bear less inventory risk and the profits from providing entertainment are split between the bookies and the “smart money”.

The bookies would be hurt the most from increased competition from other bookies, which would put downward price pressure on vig. “Smart money” is hurt most by other smart money, which would compete with them for the most profitable bets, the bets that were most easily identified as having a “line that is way out of range of reality”. In fact, with enough “smart money” there would be very few such lines that were “way out of range” (market efficiency). So, the best way to help the “dumb money” get the “fairest” betting lines would be too remove all barriers-to-entry to starting bookmaking businesses (like laws against gambling, for example) and encouraging as much smart money as possible to participate, i.e., compete against each other, in these markets.

Doug January 18, 2014 at 3:47 pm

“Betting on Amazon does change the behavior of Amazon as well as others who may or may not be betting for or against Amazon. That’s because we pretend the bet on Amazon is an investment.”

Betting on Amazon changes the company’s behavior because it affects its marginal cost of (equity) capital. As a factor of production rising or falling cost of capital changes optimal business strategy. Amazon deeply relies on the capital markets, so in turn the behavior of the speculators in that market has a direct impact on them. It does beyond just nominal conceptualization of the activity as “investing.” Even if everyone thought that stock traders were a bunch of degenerate gamblers, they’d still have enormous influence on economic activity.

MD2 January 18, 2014 at 12:02 pm

From the actual article: “He said he aims for a 2 percent return on the total amount he bets each year.”

2% return for a 60-hour work week? I assume he has to pay his trader and programmer too.

Why not work 0 hours and make 20% on the stock market in the past year?

I realize the stock market doesn’t go up 20% every year, but timing also applies to his tennis algorithm. It’s entirely possible he’s left money on the table and is building an overfitted prediction model that will cause catastrophic losses next time there’s a tennis match-fixing scandal (google it).

Charlie January 18, 2014 at 12:15 pm

You’re missing turnover. His return on bets and return on bankroll are different numbers.

Turkey Vulture January 18, 2014 at 12:37 pm

A great card-counter in blackjack, playing games with good odds and with the ability to significantly spread his bets based on count, won’t get a 2% return on his bets. The key is to be able to bet a lot. If you have a $10k bankroll but can can place $1 million in bets over the course of a year, that 2% edge gives you an expected profit of $20k. Even if he got 20% in the stock market with that $10k, that’d only be $2k.

What matters is how large his bankroll is, how much he can be in a year, how much he bets at one time, and the expected variance. I’d assume he has calculated some kind of risk-of-ruin statistic for himself, such that, say, he has a 95% chance of doubling his bankroll before going bust (given his best estimate of his edge).

Turkey Vulture January 18, 2014 at 12:39 pm

“how much he can bet in a year”

MD2 January 18, 2014 at 2:03 pm

That makes sense, thanks for clarifying. So the stock-trading equivalent would be extremely active day trading with reinvestment (as opposed to just letting a lump sum earn returns), and I guess it’s easier to be consistently profitable as an active tennis bettor than an active stock trader? Until everyone else with a computer jumps in now that they’ve read this story?

Still seems like one of those add-nothing-to-the-world occupations (unless you think tennis gambling liquidity is a really great thing). I know that wasn’t my original argument against it though, so maybe I’m just a hater.

Turkey Vulture January 18, 2014 at 4:25 pm

Yeah with stock day-trading it’s far less likely that some guy (or a guy with one or two employees) will have an edge on others in the market. There are just too many entities with more computing- and man-power similarly seeking arbitrage opportunities. But if he had some edge on the market that made the expected value of his trades +2%, it would be a similar story to this tennis deal (though of course variance would be different, and hence risk-of-ruin as well). I suspect the tennis betting market is relatively small in comparison to the stock market, or to the NFL betting market, so I can believe that some lone dude could devise a method to realize positive expected value on his bets: it hasn’t been worth another group’s time, or worth the oddsmakers’ time/money to try to produce a more perfect betting line. But it’s also possible this guy thinks he’s playing at positive expected value but has just been “lucky” so far, and that he’s really making every bet at an expected loss.

Probably adds nothing to the world, but then again, maybe some clever method used for this purpose can be applied in another realm that actually produces value for other people. That at least seems more likely with this type of thing than if someone makes a living being really good at poker.

glittering prizes January 18, 2014 at 10:29 pm

MD2 – I agree with you, anyone intelligent enough to make a middle class or better living this way is wasting his life, 60 hours at a time, this way.

Alexei Sadeski January 18, 2014 at 4:33 pm

Why is he sharing this information?

So Much For Subtlety January 18, 2014 at 6:49 pm

Because he has a proposition for you. It turns out a lot of his money is locked up in a back account in Nigeria that he cannot access ….

Steve Smith February 10, 2014 at 7:09 am

Your online sports betting post was very well written and I loved reading it. I also love to place bets and would like to read more information about this topic on your blog. The interested people who wish to bet on sports can try out my sports betting site to bet on sports and they can also get better bet to win ratios here.

Comments on this entry are closed.

Previous post:

Next post: