Syndicates, algorithms, and beating the horses in Hong Kong By Andrew Tottenham, Managing Director, Tottenham & Co November 20, 2019 at 7:00 am In news articles, we occasionally hear about betting syndicates making large bets on races and sporting events, along with accompanying expressions of concern that they bring corruption to sports. But is this true? Are syndicates a malign influence? Do they bring the risk that players will throw matches or that jockeys will pull a horse to make sure it loses? We need to be clear here that there are two types of syndicates: one a criminal enterprise that does seek to influence the outcome of matches or races so that their bets win, the other a group that uses mathematical methods to determine whether the odds offered reflect the “true” (or as near to true as possible) probabilities of the outcome of that event, and whether a bet placed has a better chance of winning than the odds would suggest. It is the latter group that I would like to discuss today. Legitimate syndicates represent millions of dollars bet at racecourses and on sporting matches around the world today. So where did syndicates come from, and how do they do what they do? In the late 1970s, two Australians, physicist Bill Benter and insurance actuary Alan Woods, learned that, by counting cards and utilizing the correct betting strategy, it was possible to win money playing blackjack in casinos. They tried it out and they won big, easily more than enough to live on. But as their notoriety spread, the number of casinos where they could continue to play diminished to the point that they found they could no longer make a living from counting cards. In the early 1980s, they decided that betting on horse races might be more promising. As we know, humans have built in biases, which is also true when it comes to betting. The odds given a particular horse to win at race time rarely reflect the true odds of that horse winning. Rather, it is the weight of money bet on each horse that influences the odds given. It turns out that, by and large, humans overestimate the chances of long shots to win, and underestimate the ability of favourites. At about the same time Benter and Woods were deciding to give up blackjack, two Canadian academics, Ruth Bolton and Randall Chapman, published a paper, “Searching for Positive Returns at the Track,” in the August 1986 edition of Management Science. They didn’t just investigate human biases; they also examined the factors that could be used to help predict the winning horse. It was probably the first paper written on the subject, and it lay the groundwork for further research. Bolton and Chapman performed an analysis to determine what they called the “quality of the horse” that used different performance factors – how fast each horse had previously run over that distance, its position at the starting gate, how heavy the horse was, etc. Using race data on North American races, collected by Chapman, the two Canadians used a technique called regression analysis, the outcome of which was to show which factors would have an impact on the result, and, in the process, proved that it was certainly possible to make this determination. But if you want to win money by betting on horses, it is not enough to know the true probabilities of the horses in the field. You also need to know whether the accuracy of your predictions can overcome the built-in edge that the bookmaker has. Bolton and Chapman’s model proved that you could use some factors and have a better than random chance of winning, but it was not sufficiently robust to overcome the house’s edge. Having read the paper, Benter realised that the problem could be due to the fact that, although Bolton and Chapman had used good race data, it was data collected at many different racetracks, so there were possibly too many variables to take into account. Benter needed a single track that held a large number of races and where the same horses tended to race each other. Luckily for him, there was one place in the world with a single track where they ran multiple races each day and which had been in operation for long enough that there was ample historic data available: the Hong Kong Jockey Club (HKJC). The HKJC offers pari-mutuel betting: the bets go into a pool, the house takes its cut, and the balance is paid out to the winners in proportion to the total amount bet. The more money bet on a winning horse, the less, proportionally, is won by the bettor. The odds (price) that the bettor receives is the last price offered before the race starts; unlike betting with traditional bookmakers, it is not possible to lock in a price from an earlier time. Benter perfected his model and, in 1994, published a paper, “Computer Based Horse Race Handicapping,” as part of an anthology called Efficiency of Racetrack Betting Markets. The model did not work profitably in practise. After losing quite a bit of money, he realised that the actual outcomes lay somewhere between what his model predicted and what the betting markets were forecasting. Benter was eventually able to tweak his model to take account of this discrepancy and bet profitably. Word spread about this new computer-based betting model, and syndicates backed mathematicians to refine the models further and make them perform even better. So many syndicates started to bet on the races at the HKJC that the market became “efficient” – in other words, the money bet reflected the true probabilities, and so there was no percentage for the syndicates to leverage. The HKJC was happy with the syndicates betting on their races because no matter what horse won, they took a cut of the amount bet. However, what it did mean was that if the syndicates made money, they did it at the expense of the “uninformed” bettor. The odds against these unfortunates were far worse that the 18% that the HKJC levies on the prize pool. Today, it could cost more than $1 million to both develop a successful computer model and acquire sufficient computing power to analyse the data and find suitable opportunities to bet, but that has not stopped the growth in the number of syndicates. Syndicates are big business; it is estimated that 20% of all the bets placed on US races today are placed on behalf of syndicates. Syndicates find it difficult to bet with traditional bookmakers. If an account wins consistently over a period of time, the account will either be closed or the size of the bet restricted. I often wonder, when I read of people complaining in chat rooms or on blogs that their accounts have been restricted, whether they are an “educated” bettor, part of a syndicate, or just lucky.