The relationship between science and betting is the focus of Adam Kucharski’s first book.

‘The Perfect Bet: How Science and Maths are Taking the Luck Out of Gambling’ has received excellent reviews from the Wall St Journal, Daily Telegraph, Financial Times, New Scientist and many others.

He’s on the podcast to talk about the evolution of scientific betting and the most successful syndicates in the world.

Punting Insights You’ll Find
– How Bill Benter won tens of millions in Hong Kong
– Why late money is smart money
– The amount of evidence they need to make a decision
– Some of the secrets behind the Computer Group in Las Vegas
– Why the Kelly Criterion is often as important as the selection model itself

More info: Adam Kucharski wrote a piece for The Guardian titled ‘Seven lucky ways that gambling changed maths’.

1. Dice games – theory of probability.

2. Problem of points – expected value.

3. Roulette and stats.

4. St Petersburg Lottery – utility.

5. Roulette – chaos theory.

6. Solitaire and the power of simulation – the Monte Carlo method.

7. Poker – game theory

Today’s Guest

Adam Kucharski

Get More Betting 360

Make sure you don’t miss our punting tips to come! Subscribe on:

or you can directly download this episode by right-clicking, Save As Here.

What Do You Think of the Show?

Podcast Transcript

David Duffield: I look forward to finding out a little bit more about the book you just published. It’s called “The Perfect Bet: How Science and Maths Are Taking the Luck Out of Gambling”.

I haven’t had the chance to read it just yet, it’s only just come out, but you must be pretty excited about some of the reviews from the Wall Street Journal, and the Daily Telegraph, and Financial Times and a few others.

Adam Kucharski: Yeah, it’s great to see some attention picking up. I think a lot of people have an interest in this kind of thing, looking at risk. And from different angles. You know the Wall Street Journal are obviously focused on finance. I think in the U.K. as well there’s just a lot of people who are interested in these games and looking at ways that you can find gaps and loopholes.

David Duffield: The book discusses how science has influenced gambling and also the other way around. Tell us a bit about the evolution of scientific betting in both sports and horse racing.

Adam Kucharski: One surprising thing I found is really going, as soon as you go back in history, a lot of ideas we use in maths originated with these kind of games. A lot of probability and statistics, even a few hundred years ago, came from people trying to study games of chance. The ideas now have really kind of gone the other way. So, people have kind of taken ideas from science, taken ideas from gambling, and used them to kind of take on things like horse racing and sports betting.

One of the people I talked to in the researching the book was Bill Benter, who was very successful in Hong Kong in horse racing. One of the surprising things, of course, is that he didn’t actually start his career in horse racing or in sports. It was actually in blackjack that he would find a lot of ideas.

Blackjack was one of these classic examples of a casino game which was beaten with math. On the face of it, blackjack is something that’s random. You’ve got these deals coming out, and you’ve got to make it to twenty-one, but there’s not very much control you can do about it. The casino has an edge so you can adapt your tactics a little bit depending on what the dealer’s cards is showing, but it’s pretty hard to get beyond that. Card counting has the advantage that you can keep track of what’s come before, so if you know what cards have already been dealt that gives you a sense of what’s going to come next. By doing that effectively, you can get an edge on the house. That’s what Bill Benter and a lot of other people were doing in the late 70s.

What the challenge of obviously betting in the casinos is security very quickly clamp down on you. As one guy I talked to said, it’s very easy to learn to card count, but it’s very hard to learn how to get away with it. So what a lot of those gamblers did who were initially in casinos, moved on to other games. For a couple of them, horse racing presented a natural target.

The reason, actually, that they pick Hong Kong as a location for this kind of task is actually from a scientific point of view in what’s an ideal laboratory. You’ve got basically a small number of horses, about a thousand or so horses, racing on two tracks and racing against each other repeatedly. So, if you think of somewhere like the U.S. for instance, there’s a huge number of different horse racing tracks, tons of different horses; very rarely are you racing against each other. So, it’s quite hard to come up with a consistent data set that you can analyse. Hong Kong has that nice property that you just get these horses racing against each other again and again.

Of course, you need a way to measure, given that data, of how good each horse’s performance is. Actually, it was visit to a library, Bill Benter used in Nevada, he came across an article that had been published in the Journal of Management. It was essentially taking the statistical model and converting all of this information, horse performances, whether it was number of races run, average time, all these different factors and converting into a measure of performance. This is essentially the framework used in betting in Hong Kong, they subsequently refined this approach over time. In the first season, they took a pretty big hit. They lost a few thousand dollars, but in subsequent seasons, as they got more data and kind of refined it down, they actually got something that was performing a lot better than the market.

So, in the Hong Kong you’ve got parimutuel markets so the odds on display completely depend on what other people are betting on. The more people that bet on a certain horse, the lower this horse’s odds are going to be. So really it’s a matter of coming up with a model and a prediction which can outperform the public. That’s really been one of the early big developments in terms of betting on these kinds of markets. People just found the correct laboratory to test their ideas and those guys have been incredibly successful. They’ve wagered probably tens if not hundreds of millions in years and they’ve made phenomenal profit. Almost to the point where these kind of syndicates in Hong Kong, don’t actually celebrate their wins. It’s really, for them, more unusual for them to be losing consistently because they’ve just got this framework set up where they want to make consistent bets and consistent profits.

David Duffield: So you mentioned that Bill learned a lot of techniques from blackjack. I think people will understand the card counting aspect and how that can be turned in your favour, to put the percentages in your favour. What do you think the parallels were between blackjack and horse racing?

Adam Kucharski: I think one of the big parallels is kind of the question you’re asking. I think with casino games in particular, there’s a lot of systems and a lot of superstitions around them. One of the things that I found quite interesting talking to these syndicates is the evidence they need to make a decision and what kind of questions are important to them when they’re making these betting decisions. Actually, in some of the early models Bill Benter and that syndicate put together, certain factors came out as very important when shaping horses’ performance. I think the number of races a horse has run previously is one of them. It’s kind of tempting to come up with a narrative around that. You want to say, well if they’ve run more races that’s going to give them more experience, that’s going to give them better performance. But really, they realised it’s a very complex situation. They’ve got a huge amount of data going through these models, everything kind of overlaps a bit. It’s not quite clear that there’s one single thing that they can pluck out as being important and explain why.

For them, actually they have very little interest in explaining why a horse has won. Really, what they want is to know what horse is going to win. So, as long as they’ve got a model which performs very well against past data and gives them very good predictions and better predictions than the markets, they don’t actually bother too much about interpreting them. They don’t actually mind they’re not becoming experts at horse racing, because their interest is to get the correct horse and get something that can win rather than having this real deep understanding as to why particular factors may come out as important in their analysis.

David Duffield: So there was Bill who is fairly well known and very successful, but you also spent some time with the people from the Computer Group in Vegas?

Adam Kucharski: Talking to some of the people who started out with Michael Kent, who was one of the people who worked a lot in computer models. Again, for him, this was an interest that started much further back. He was saying that even in high school he would, when he was a bit bored in class, he’d be looking through newspapers and looking at scores and saying, “If someone beats somebody by a certain amount, how good is that?” “How do you actually measure how good a performance that is?”

Actually, for him, it was a series of almost fortunate timings that led him into that industry because this was in the 70s and he was working for Westinghouse Corporations. This is an atomic power group designing things for the Navy. It was here that he learned a lot about computing, a lot about programming and about how to develop these models of a process. In this case, he was working in science. But in his spare time, he started developing these college sport models. It was very similar to the approach that the horse racing teams were using. You’d need to gather together a lot of this data and try to work out which factors are important in shaping these results.

Obviously, U.S. sport has quite a long history in people using data. Baseball is the classic one with sabermetrics that you collect together, huge amounts of stuff … Just the way sports are structured in the U.S., there’s a much bigger role that statistics plays, but other things as well were beneficial.

Michael Kent was based in Pittsburgh, and they had a library of sports results for college sports, college basketball, college football, that sort of thing. It was fantastic resource, but he had to manually input these results into his computer. He was in a quite fortunate position that he had this experience and these resources, but it required a pretty amazing amount of work to get it set up. But once he had this ready, it actually gave him quite a large edge because this was a point where, in Vegas at least, a lot of the betting was really not done on a scientific basis. So having this model where you could make pretty objective measures of how teams were going to do gave you kind of a valuable input into this.

It was interesting as well just talking about the logistics of this, Kent teamed up with Billy Walters, another well known gambler in the U.S. Walters was much more familiar with Vegas and the setup and actually execution of how you go about carrying out these bets. That’s one of the reasons these teams actually worked well together.

It’s really valuable in Hong Kong actually. You can make bets over the phone, which is different to other places and if you’re trying to get bets on in the last few minutes and make bets on changing odds having that quick access can make a very big difference.

David Duffield: So, there’s also some discussions you had on the football side of things, I suppose soccer in our terminology down here, but tell us about the Dixon and Coles method.

Adam Kucharski: Yes. Again, soccer seems to be one of those things that came along a bit later relative to other sports. Although in the US there’s this big history of looking at baseball and the Computer Group were targeting college football and basketball. Soccer seems one of those things that’s far more challenging to actually predict. Actually, in the 70s and 80s, a few researchers were looking at a game and looking at the stats, typically it’s very low scoring. They came to the conclusion that it’s not possible to make reliable predictions. This is something that’s influenced by chance. Although, obviously over the course of the league, the better team might come out on top, in a single game there’s so much potential for an upset so it’s very difficult to predict.

Actually, in the 1990s in Lancaster University in the north of the U.K up by Manchester, there were a couple of university lecturers and that year, a university exam, a final exam for statistics students, someone had dropped in a question about predicting soccer. It was a bit of a simplistic question, but when these lecturers found it, these two lecturers Stuart Coles and Mark Dixon, they realized that there’s a bit of potential, Mark Dixon in particular was interested in working out how this method could be improved.

Over time, they got hold of some data and started to refine thir model, almost as a side project. And realised that they had something which not only could come up with decent predictions, could actually outperform bookmakers.

In terms of developing the model, one of the key things they came up with is up until that point a lot of these predictive models for soccer treated all teams as the same. In other words, they tried to work out what’s the average goal a team might score, if two teams played each other therefore there’s a certain probability of obtaining a certain result. What Dixon and Coles did, and actually pretty much all syndicates who bet on soccer use a similar approach or at least that’s their foundation now, is looked at higher dimensions. Rather than just having a single value for goals, or even just a single value for each team, so in the league each team is ranked by a certain value, they actually had multiple characteristics for the team. In their model they had each team an attacking potential and a defensive potential, as well as a term that controlled for the home advantage. If you think about it, you have two teams that are very similar, but depending on whether they’re playing home or away they can get very different results.

Having that additional detail in the model enabled them to reproduce some of these quirks of soccer matches and actually come up with something a lot better than what the bookmakers were coming up with. Over time, Mark Dixon actually set up ATASS, which is a statistical sports consultancy. Then a few years later, Stuart Coles joined Smartodds, which is another company focused on the sports predictions models.

Just in the last decade or two, in the UK there’s been a huge rise in these companies who focus on these kind of things. Starlizard is another one which has picked up a bit of attention recently. I think a lot of these developments are relatively recent compared to what’s happened in the U.S. just because these team sports like soccer, which are low scoring, are generally much harder to predict. You need that availability of data and development of new methods to actually make that possible. So, I think that it’s quite interesting, it’s quite exciting to see a sport that’s traditionally has been seen as one that’s very difficult to predict actually come to the point where you’ve got syndicates making pretty good consistent money on them by analysing what’s going on. A lot of the methods are now feeding into team management in soccer and to pretty much every team that the Premier League now has a team statisticians working on crunching through all the data that’s become available and in their case using it to make decisions about strategy and player signings.

So, I think there’s a bit of a shift happening that potentially has happened a while ago. Moneyball and this sort of thing in the U.S. is now actually coming into other sports. It’ll be interesting to see how it all develops.

David Duffield: So whether its Bill Banter in Hong Kong, Computer Group in Vegas, all the big boys in England like Starlizard or Smartodds, have you found that the market has become more and more efficient each year or that they have smaller margins, basically?

Adam Kucharski: In some areas, definitely. Talking to some of the guys in Hong Kong, they say it’s pretty brutal now because you’ve got so many bright guys who’ve gone there and tried to bet together. I think as a result, people have really come looking to other markets. I think in the U.S. in particular, some people now in a lot of computer teams at those race tracks. Likewise in the U.K., people are kind of moving on to other sports. Things potentially like golf or tennis tends to be a bit trickier, either because there’s just a lack of data or because there’s just so much uncertainty around a number of competitors in those certain events.

I think that’s always been the case with the successful betting syndicates. It’s matter of looking where people aren’t putting their focus on.

I think in the research of the book and actually talking to all these guys who developed these different ideas over the last few decades, it very much seems it’s the people, almost the outsiders who are not in an industry or in a particular sport at the time, coming in with some new ideas, new methods and doing a lot better than everyone else is.

David Duffield: You met some pretty successful and pretty interesting people in writing and researching the book. Is there any advice you’d give my listening audience? We’re all punters, we all love a bet. Is there anything you picked up that were similar traits amongst all these successful people?

Adam Kucharski: I think one of the main things that jumps out here is the pretty scientific approach that all these guys are using. From Hong Kong to the guys in the UK, a lot of them have really thought about these things in an objective way. I think there’s a huge amount of stories in gambling and everyone has a system that they’re convinced works. I think it’s almost been the ability of most of these people to embrace their ignorance. Michael Kent, when I was talking to him was saying that it’s something that you constantly have to test. You have to build your model, you have to test it , you have to update it and that’s a constant process. You cant really stick with something that you think wins.

Another thing that really jumped out was the amount of focus on the execution. Things like the money management and ideas like the Kelly Criterion, which dictates how much you should risk. If you’ve got a certain edge on a bet, how much a bankroll should you put on that bet. Actually, for many of these teams, that’s a really important component of their betting strategies. It’s not just about coming up with these predictions, it’s actually how you convert that into something that can win. In some cases, in Hong Kong for instance, the predictions they were coming up with, so these kind of predictions made in a vacuum away from everyone else, benefited by also accounting a little bit for what was going on in the market. With something like horse racing, potentially there are people with privileged information. They might know something about the jockey’s routine, or the training or this sort of thing. So actually, once you’ve come up with the prediction, then comparing it to the market and seeing if there’s additional information there. It’s really about how you’re combining this pure scientific approach with what’s going on reality and finding the best way to actually come up with a profitable strategy.

David Duffield: Did anyone go into detail on that? Because the interesting thing about the Kelly Criterion, the theory is that the bigger your edge (or your perceived edge), the more you should bet. But then there’s also what you mentioned before about the wisdom of crowds and market intelligence, which would say if you’ve rated something really short but the market’s giving you a massive price, you might want to temper your bet a bit. How did the syndicates handle that?

Adam Kucharski: One of the things, as you mentioned, with Kelly that kind of comes up with a nice theoretical value if you know what your edge is and that sort of thing. Any may cases, syndicates will bet a fraction of Kelly because they don’t fully trust that their predication of what their edge is, is completely correct. Obviously, if that’s off, it’s going to really affect your chances. In something like a parimutuel market as well one of
he issues that they had is bets would often come in pretty late, so you might see a few hundred thousand on the board and then by the time you get your bets in at a certain price, a load of money would come in on another horse, or worse, the same horse and skew the odds.

I think, as Bill Banter said, late money is generally the smart money. So, often they adjust their Kelly bets and then their predictions. That almost takes it the next level. A lot of the focus, and similarly for a lot of the modern soccer betting syndicates, that a lot of the effort is put into avoiding moving markets. If you’ve got competitors that want to know what you’re gonna do and want to know where you’re placing your bets, then you want ways of actually moving that money into the market in a clever way that means that other people can’t capitalise on what you’re up to. And in the case of Hong Kong, but even if people come in with that later or, you’ve got an idea of where those bets are going to go and how big they’re going to be if you can adjust your strategy accordingly.

David Duffield: Well I really appreciate the chat, Adam. I haven’t been able to read the book yet, but it’s on my list and I look forward to doing that and we’ll chat again soon.

Adam Kucharski: Yeah, sounds good. Thank you.

LEAVE A REPLY