The Perfect Bet Read online

  The “Sports Illustrated jinx” is a well-known example of this problem: often a player who appears on the cover of Sports Illustrated subsequently suffers a dip in form. Statisticians have pointed out that the Sports Illustrated jinx is not really a jinx. Players who end up on the cover often do so because they’ve had an unusually good season, which was down to random variation rather than a reflection of their true ability. The drop in performance that came the following year was simply a case of regression to the mean, just as Francis Galton found while studying inheritance.

  When a club signs a new player, it has to make decisions based on past accomplishments. Yet what it is really paying for is future performances. How can a sports club predict a player’s true ability? Ideally, it would be possible to pull past performances apart and work out how much they were influenced by ability and by chance. Statistician James Albert has attempted to do this for baseball. By trawling through lots of different statistics for pitchers, including wins and losses, strikeouts—where the batter misses the ball three times—and runs scored against them. He found that the number of strikeouts was the most accurate representation of a pitcher’s true skill, whereas statistics such as home runs conceded were more subject to chance, and hence are a poor reflection of pitching ability.

  Other sports are trickier to analyze. Soccer pundits generally use simple measurements, such as goals per game, to quantify how good strikers are. But what if strikers play for a good team and benefit from having other players setting them up with scoring chances? In 2014, researchers at Smartodds and the University of Salford assessed the goal-scoring ability of different soccer players. Rather than just asking how likely a striker was to score, they split goal scoring into two components: the process of generating a shot—which could be influenced by the team performance—and the process of converting that shot into a goal. Splitting up scoring in this way led to far better predictions about future goal tallies than simple goals-per-game statistics provided. The study also produced some unexpected conclusions. For instance, it appeared that the number of shots a player had bore little relation to the team’s attacking ability. In other words, good players generally end up with a similar shot count regardless of whether they are playing for a great team or a weak one. Although better teams have more shots overall, a decent player ends up being a little fish in a large scoring pool; at a struggling club, that same player can make a bigger contribution to the total. The researchers also found that it was difficult to predict how often a player would convert shots into goals. Hence, they suggest that team managers looking at a potential signing should estimate how many shots that player generates rather than how many goals the player scores.

  WHEN IT COMES TO scientific sports betting, the most successful gamblers are often the ones who study games others have neglected. From Michael Kent’s work on college football to Mark Dixon and Stuart Coles’s research in soccer, the big money generally comes from moving away from what everyone is doing.

  Over time, bookmakers and gamblers have gradually latched on to the best-known strategies. As a result, it is becoming harder to profit from major sports leagues. Erroneous odds are less common, and competitors are quick to jump on any advantage. New syndicates are therefore better off focusing their attention on lesser-known sports, where scientific ideas have often been ignored. According to Haralabos Voulgaris, this is where the biggest opportunities lie. “I would start with the minor sports,” he said at the MIT Sloan Sports Analytics Conference in 2013. “College basketball, golf, NASCAR, tennis.”

  In minority sports, additional knowledge—whether from models or experts—can prove extremely valuable. Because crucial variables are not so well known, the difference in skill between a sharp bettor and a casual gambler can be huge. As well as helping gamblers build better predictive models, improvements in technology are also changing how bets are made. The days of suitcases full of banknotes are coming to an end. Bets can be placed online, and gamblers can control hundreds of wagers at the same time. This technology has also paved the way for new types of strategy. A large part of sports betting throughout its history has been about forecasting the correct result. But scientific betting is no longer just a matter of predicting score lines. In some cases, it is becoming possible to know nothing about the result and still make money.

  5

  RISE OF THE ROBOTS

  “WHAT HATH GOD WROUGHT!” THE MESSAGE READ. IT WAS May 24, 1844, and the world’s first long-distance telegram had just arrived in Baltimore. Thanks to Samuel Morse’s new telegraph machine, the biblical quote had traveled along a wire all the way from Washington, DC. Over the next few years, single-wire telegraph systems spread around the globe, creeping into the heart of all sorts of industries. Railway companies used them to send signals between stations, while police fired off telegrams to get ahead of fleeing criminals. It wasn’t long before British financiers got ahold of the telegraph, too, and realized that it could be a new way to make money.

  At the time, stock exchanges in the United Kingdom operated independently in each region. This meant there were occasional differences in prices. For example, it was sometimes possible to buy a stock for one price in London and sell it for a higher price in one of the provinces. Obtain such information quickly enough, and there was a profit to be made. During the 1850s, traders used telegrams to tell each other about discrepancies, cashing in on the difference before the price changed. From 1866 onward, America and Europe were linked by a transatlantic cable, which meant traders were able to spot incorrect prices even faster. The messages that traveled down the wire were to become an important part of finance (even today, traders refer to the GBP/USD exchange rate as “cable”).

  The invention of the telegraph meant that if prices were out of line in two locations, traders had the means to take advantage of the situation by buying at the cheaper price and selling at the higher one. In economics, the technique is known as “arbitrage.” Even before the invention of the telegraph, so-called arbitrageurs had been on the hunt for mismatched prices. In the seventeenth century, English goldsmiths would melt down silver coins if the price of silver climbed past the value of the coin. Some would even trek further afield, hauling gold from London to Amsterdam to capitalize on differences in the rate of exchange.

  Arbitrage can also work in gambling. Bookmakers and betting exchanges are merely different markets trading the same thing. They all have varying levels of betting activity and contrasting opinions about what might happen, which means their odds won’t necessarily line up. The trick is to find a combination of bets so that whatever happens, the payoff will be positive. Suppose you’re watching a tennis match between Rafael Nadal and Novak Djokovic. If one bookmaker is offering odds of 2.1 on Nadal, and another is offering 2.1 on Djokovic, betting $100 on each player will net you $210—and cost you $100—whatever the result. Whoever wins, you walk away with a profit of $10.

  Unlike syndicates working on sports prediction, which are in essence betting that their forecast is closer to the truth than the odds suggest, arbitrageurs don’t need to take a view on what will happen. Whatever the result, the strategy should lead to a guaranteed profit, so long as a gambler can spot the opportunity in the first place. But how common are arbitrage situations?

  In 2008, researchers at Athens University looked at bookmakers’ odds on 12,420 different soccer matches in Europe and found 63 arbitrage opportunities. Most of the discrepancies occurred during competitions such as the European Championship. This was not particularly surprising, because tournament results are generally more variable than results in leagues where teams play each other often.

  The following year, a group at the University of Zurich searched for potential arbitrage in odds given by betting exchanges like Betfair as well as traditional bookmakers. When they considered both types of market, there were far more stray odds. They found it would have been possible to make a guaranteed profit on almost a quarter of games. The average return wasn’t huge—around 1 to
2 percent per game—but it was clear there were enough inconsistencies to make arbitrage a viable option.

  Despite the allure of arbitrage betting, there are some potential pitfalls. To be successful, gamblers need to set up accounts with a large number of bookmakers. These companies usually make it easy to deposit money but hard to withdraw it. Bets also need to be placed simultaneously: if one wager lags behind another, the odds might change, thwarting any chance of a guaranteed profit. Even if gamblers can overcome these logistical issues, they have to avoid attracting the attention of the bookmakers themselves, who generally dislike having arbitrageurs cutting into their profits.

  It is not just differences between bookmakers that can be exploited. Economist Milton Friedman pointed out that there is a paradox when it comes to trading. Markets need arbitrageurs to take advantage of incorrect prices and make them more efficient. Yet, by definition, an efficient market shouldn’t be exploitable, and hence shouldn’t attract arbitrageurs. How can we explain this contradictory situation? In reality, it turns out that markets often have short-term inefficiencies. There are periods of time when prices (or betting odds) do not reflect what is really going on. Although the information is out there, it hasn’t been processed properly yet.

  After a major event—such as a goal being scored—gamblers on betting exchanges need to update their opinions on what the odds should be. During this period of uncertainty, whoever reacts to the news first will be able to place bets against opponents who have not yet adjusted their odds. There is a limited window in which to do this. Over time the market will become more efficient, and the available odds will change to reflect the new information. In 2008, a group of researchers at the University of Lancaster reported that it takes less than sixty seconds for gamblers on betting exchanges to adjust to a dramatic event in a football match.

  Not only is the betting window small, potential gains can be modest, too. To profit a gambler would need to place a large number of bets, and place them quickly. Unfortunately, this is not something that humans are particularly good at. We take time to process information. We hesitate. We struggle with multiple tasks. As a result, some gamblers are choosing to step back from the bustle of hectic betting markets. Where humans falter, the robots are rising.

  THERE ARE TWO WAYS to access the Betfair betting exchange. Most people simply go to the website, which displays the latest odds as they become available. But there is another option. Gamblers can also bypass the website and link their computers directly to the exchange. This makes it possible to write computer programs that can place bets automatically. These robot gamblers have plenty of advantages over humans: they are faster, more focused, and they can bet on dozens of games at once. The speed of betting exchanges also works in their favor. Betfair is quick to pair up people who want to bet for a particular event with those intending to bet against it. Of the 4.4 million bets placed on the day of England’s opening match in the 2006 soccer World Cup, all but twenty were handled in less than a second.

  Automated gamblers are increasingly common in betting. According to sports analyst David Hastie, there are plenty of bots out there searching for stray odds and exploiting other gamblers’ mistakes. “These algorithms mop up any mispricing,” he said. The presence of artificial arbitrageurs makes it difficult for humans to cash in on such opportunities. Even if they spot an erroneous price, it’s often too late to do anything about it. The bots will already be placing bets, removing these slices of profit from the market.

  Arbitrage algorithms are also becoming popular in finance. As in betting, the faster the better. Companies are doing all they can to ensure they get to the action before their competitors do. It has led to many firms placing their computers directly next to stock exchange servers. When the market reacts quickly, even a slightly longer wire can lead to a critical delay in making a trade.

  Some are going to even more extreme lengths. In 2011, US firm Hibernia Atlantic started work on a new $300 million transatlantic cable, which will allow data to cross the ocean faster than ever before. Unlike previous wires, it will be directly below the flight path from New York to London, the shortest possible route between the cities. It currently takes 65 milliseconds for messages to travel the Atlantic; the new cable aims to cut that down to 59. To give a sense of the scales involved, one blink of the human eye takes 300 milliseconds.

  Fast trading algorithms are helping firms learn about new events first and act on them before others do. Yet not all bots are chasing arbitrage opportunities. In fact, some have the opposite aim. While arbitrage algorithms are searching for lucrative information, other bots are trying to conceal it.

  WHEN SYNDICATES BET ON horse races in Hong Kong, they know that the odds will change after they’ve placed their bets. This is because in pari-mutuel betting the odds depend on the size of the betting pool. Teams therefore have to account for the shift when developing a betting strategy. If they put down too much, and shift the odds too far, they might end up worse off than if they’d bet less.

  The problem also appears in sports betting. If you try to put down a large amount of money on a football match, it will be the bookmakers—or betting exchange users—who move the odds against you. Let’s say you want to bet $500,000 on a certain outcome. One bookmaker might offer you odds that would return double the stake. But the bookmaker might only be willing to take a bet of $100,000 at those odds. After you place that initial bet, the bookmaker’s odds will probably drop. Which means you’ve still got $400,000 you want to bet, and you’ve already disrupted the market. So, if you bet another $100,000 at the new odds, you won’t quite double your money. You might get even lower odds for the next chunk of cash, and things will continue to get worse with each bet you make.

  Traders call the problem “slippage.” Although the price initially on offer might look good, it can slip to a less favorable price as the transaction is made. How can you get around the problem? Well, you could try to hunt down a bookmaker who’ll take the bet in one go. At best, this could take a while; at worst, you’ll never find one. Alternatively, you could place the first bet of $100,000 and then wait and hope the bookmaker’s odds will rise again so you can bet the next chunk of money. Which is clearly not the most reliable strategy either.

  A better approach would be to mimic the tactics employed by betting exchanges. Betfair’s early success was in part the result of the way it handled each bet. Rather than attempting to find a gambler who wanted to accept a bet of the exact same size, Betfair sliced up the bet into smaller chunks. It was far easier—and quicker—to find several users happy to take on these little wagers than to hunt down a single gambler willing to take the whole bet.

  The same idea makes it possible to sneak a trade into the market with limited slippage. Instead of trying to offload the whole trade at once, so-called order-routing algorithms can slice the main trade up into a series of smaller “child” orders, which can easily be completed. For the process to work effectively, algorithms need to have good knowledge of the market. As well as having information on who’s happy to take the other side of each trade—and at what price—the program has to time the transactions carefully to reduce the chances of the market moving before the trade is complete. The resulting trade is known as an “iceberg order”: although competitors see small amounts of trading activity, they never know what the full transaction looks like. After all, traders don’t want rivals shifting prices because they know a big order is about to arrive. Nor do they want others to know what their trading strategy is.

  Because such information is valuable, some competitors employ programs that can search for iceberg trades. One example is a “sniffing algorithm,” which make lots of little trades to try to detect the presence of big orders. After the sniffing program submits each trade, it measures how quickly it takes to get snapped up in the market. If there’s a big order lurking somewhere, the trades might go through faster. It’s a bit like dropping coins into a well and listening for the splashes to work out ho
w deep it is.

  Although bots allow gamblers and banks to carry out multiple transactions quickly, they do not always act in the interests of their owners. Left unsupervised, bots can behave in unexpected ways. And sometimes they wander deep into trouble.

  BY THE TIME THE 2011 Christmas Hurdle at Dublin’s Leopardstown Racecourse reached the halfway mark, the race was as good as won. It was just after two o’clock, and the horse named Voler La Vedette was already leading by a good distance. As the hooves pounded the ground on that cold December afternoon, nobody with any sense would have bet against that horse.

  Yet somebody did. Even as Voler La Vedette approached the line, the Betfair online market was displaying extremely favorable odds for the horse that was almost certain to win. It appeared that someone was happy to accept bets at odds of 28: for every £1 bet, the bettor was offering to pay £28 if the horse won. Very happy, in fact. This remarkably pessimistic gambler was offering to accept £21 million worth of bets. If Voler La Vedette came first, the gambler would be on the hook for almost £600 million.

  Soon after the race finished, one Betfair user posted a message on the website’s forum. Having witnessed the whole bizarre situation, the user joked that someone must have been giving bettors a Christmas bonus. Others chipped in with potential explanations for the mishap. Maybe a gambler had suffered an attack of “fat fingers” and hit the wrong number on the keyboard?

  It didn’t take long for another user to suggest what might really have been going on. The person had noticed something odd about that offer to match £21 million of bets. To be precise, the number displayed on the exchange was just under £21.5 million. The user pointed out that computer programs often store binary data in units that contain thirty-two values, known as “bits.” So, if the rogue gambler had designed a 32-bit program to bet automatically, the largest positive number the bot would be able to input on the exchange would be 2,147,483,648 pence. Which meant that if the bot had been doubling up its bets—just as misguided Parisian gamblers used to do while betting on roulette in the eighteenth century—£21.5 million is the highest it would have been able to go.