WEBVTT - Episode 13: How a Professor Won Gambling on an Obscure Sport 0:00:04.880 --> 0:00:08.119 Hello, and welcome to another episode of Odd Lots. I'm 0:00:08.200 --> 0:00:11.920 Joe Wisenthal, Managing editor of Bloomberg Markets, and I'm Tracy Alloway, 0:00:11.960 --> 0:00:15.280 Executive editor of Bloomberg Markets. Hey, Tracy, do you know 0:00:15.400 --> 0:00:19.320 what the fastest sport in the world is? Uh? Sailing? 0:00:19.720 --> 0:00:23.439 Race car driving? Uh, those are pretty good guesses, but no, 0:00:23.760 --> 0:00:28.400 Actually the answer is high lie high lie highlight. Highlight 0:00:28.600 --> 0:00:34.080 high Okay, highlight. It's actually originally a game that originated 0:00:34.159 --> 0:00:38.040 in the Spanish Basque country. It's kind of like racquetball, 0:00:38.159 --> 0:00:41.600 except the players play on this gigantic court. The ball 0:00:41.720 --> 0:00:46.240 goes nearly two miles the rackets of these gigantic crald 0:00:46.280 --> 0:00:50.000 things that the players wear over their hands. The ball 0:00:50.120 --> 0:00:52.640 is hard as a rock, and oh yeah, if it 0:00:52.680 --> 0:00:54.440 were to hit you in the head, the ball could 0:00:54.520 --> 0:01:00.000 kill you. This sounds like a made up sport. Um, 0:01:00.080 --> 0:01:02.480 why are we talking about this? It's a good question because, 0:01:02.520 --> 0:01:05.920 in addition to how crazy and intense the game is, 0:01:06.440 --> 0:01:08.280 you know, as I said, two hundred mile an hour 0:01:08.920 --> 0:01:12.520 ball is hard as a golf ball, potentially deadly. People 0:01:12.560 --> 0:01:16.280 actually gamble on highlight, kind of like horse racing. Uh. 0:01:16.560 --> 0:01:19.120 The people a bunch of players play in a tournament 0:01:19.200 --> 0:01:22.080 of sorts and then people bet on whether the different 0:01:22.080 --> 0:01:25.119 players will win place or shows. So there's this big 0:01:25.160 --> 0:01:28.920 gambling element to it that's really interesting. Our guest today 0:01:28.920 --> 0:01:31.360 that we'll be talking to is Steven Skina. He's a 0:01:31.400 --> 0:01:34.400 professor at stony Brook University and he wrote a book 0:01:34.440 --> 0:01:38.440 all about gambling on Highlight Online and how he cracked 0:01:38.480 --> 0:01:40.680 the system and he made a bunch of money. He 0:01:40.680 --> 0:01:43.000 cracked the system, so he beat the house. Yeah, it's 0:01:43.000 --> 0:01:47.080 basically impossible theoretically, you know, gambling, the house is always 0:01:47.120 --> 0:01:49.920 supposed to lose. But in these games we're sort of 0:01:49.920 --> 0:01:52.920 betting against other people and the crowd sets the odds. 0:01:53.040 --> 0:01:56.400 It's actually possible. And not only did Steven beat the 0:01:56.400 --> 0:01:58.880 system and make a bunch of money, there's some interesting 0:01:58.920 --> 0:02:01.760 lessons in terms of eating the stock market and odds 0:02:01.800 --> 0:02:04.960 games in general. Okay, so I'm excited because not only 0:02:05.000 --> 0:02:07.840 am I about to learn about a sport which I've 0:02:07.880 --> 0:02:10.960 never heard of before, but I'm also interested in making money. 0:02:11.040 --> 0:02:19.520 So this sounds too all right, I think you've seen 0:02:19.560 --> 0:02:25.040 it all well, think fast, experience the wall to wall 0:02:25.080 --> 0:02:29.400 action and NonStop excitement that is Miamie Highlive see world 0:02:29.440 --> 0:02:32.160 class athletes climbed the walls to catch a rock hard 0:02:32.200 --> 0:02:35.600 ball flying. It speeds over a hundreds of miles an hour. 0:02:36.240 --> 0:02:44.200 Think excitement, Think Miami, Higlan, I think fast Steven, thank 0:02:44.240 --> 0:02:46.000 you very much for joining us. Thank you, it's nice 0:02:46.040 --> 0:02:49.079 to be here. Tracy has never heard of highlight before. 0:02:49.600 --> 0:02:52.360 What is highlight and how did you get interested in? So? 0:02:52.480 --> 0:02:55.880 Highlight is as as you said, a besque game. Originally, 0:02:55.960 --> 0:02:59.120 it's sort of like a variation on handball. Um. The 0:02:59.160 --> 0:03:01.600 reason people are exposed to it in the United States 0:03:01.639 --> 0:03:06.160 is typically because in Florida it's a betting venue. Um. 0:03:06.200 --> 0:03:09.960 There are these stadiums called front Times in Miami and 0:03:10.080 --> 0:03:15.240 Dania near Fort Lauderdale where you can watch mostly basque 0:03:15.840 --> 0:03:19.440 players play the sport. And that's actually exactly how I 0:03:19.520 --> 0:03:21.960 discovered it. My family used to go down to Florida, 0:03:22.200 --> 0:03:25.800 my North Miami beach every winter for a couple of weeks, 0:03:26.120 --> 0:03:29.639 and we went occasionally to UH the Dania Highlight front Time. 0:03:29.800 --> 0:03:32.040 That was exactly the way that we got involved. You know, 0:03:32.040 --> 0:03:34.679 our family would every year take its vacation visiting the 0:03:34.760 --> 0:03:38.040 relatives in Florida. We would drive down and we would 0:03:38.200 --> 0:03:41.040 UH one night go to the Highlight front time. Is 0:03:41.040 --> 0:03:44.280 it fun to watch these games? It sounds intense from 0:03:44.320 --> 0:03:47.960 Joe's description, it's I think it's incredibly exciting. I mean, 0:03:48.000 --> 0:03:50.200 first of all, it's fun to watch them make these 0:03:50.200 --> 0:03:52.800 plays because the ball is moving very fast, they have 0:03:52.920 --> 0:03:56.320 to make, you know, great catches and very difficult throws. 0:03:56.360 --> 0:03:59.560 But also the scoring system involved in Highlight has some 0:03:59.680 --> 0:04:03.040 infra sting mathematics instructor that makes it kind of fun 0:04:03.080 --> 0:04:07.080 to watch. So depending up on how you bet and um, 0:04:07.200 --> 0:04:10.760 the chances of winning change extremely rapidly with every point, 0:04:11.080 --> 0:04:13.480 and so it's it's it's very exciting because the situation 0:04:13.560 --> 0:04:16.240 is always changing. Yeah, so I said it was kind 0:04:16.320 --> 0:04:19.120 of like horse race betting and that you could bet 0:04:19.160 --> 0:04:20.840 on wind, place their show. But in a way it's 0:04:20.839 --> 0:04:23.159 a little more complicated. Why don't you just sort of 0:04:23.200 --> 0:04:26.640 describe how the similarities real quickly and the differences. So 0:04:26.720 --> 0:04:28.920 it is like horse race betting and that you can 0:04:28.960 --> 0:04:32.040 bet on win, place and show, and that's certainly what um, 0:04:32.080 --> 0:04:36.719 what we're gonna be doing. The difference is the scoring system. UM. Basically, 0:04:37.320 --> 0:04:39.840 in Highlight the winner is the first one to get 0:04:39.880 --> 0:04:42.640 the seven points, and they have eight teams that are 0:04:42.640 --> 0:04:45.640 playing in any given match, but because of the size 0:04:45.680 --> 0:04:47.880 of the court, only two teams can play at once. 0:04:48.520 --> 0:04:51.640 The teams wait in a line. Their their uniform numbers 0:04:51.640 --> 0:04:54.760 are one through weight, corresponding to where they start in line. 0:04:55.440 --> 0:04:59.320 And originally the first two players play each other. Then 0:04:59.360 --> 0:05:02.520 the winner keeps playing, gets a point and keeps playing. 0:05:02.520 --> 0:05:05.240 The loser goes to the end of the line, and 0:05:06.120 --> 0:05:09.159 they keep playing until you get to seven points. And 0:05:09.200 --> 0:05:11.600 if you think about that kind of a scoring system, 0:05:11.640 --> 0:05:14.160 it gives an advantage to the people who start early 0:05:14.920 --> 0:05:18.640 because um, obviously they get first cracks at getting points, 0:05:19.200 --> 0:05:21.680 and even if they lose, they are more likely to 0:05:21.720 --> 0:05:23.800 be they're gonna be the first player to come up 0:05:23.839 --> 0:05:27.320 for a second time. So they make the scorches them 0:05:27.360 --> 0:05:31.240 even more complicated. Whereafter every trip through the queue once 0:05:31.480 --> 0:05:34.080 meaning every player has played its first point, now every 0:05:34.120 --> 0:05:37.640 subsequent point counts for two. And this makes for a 0:05:37.720 --> 0:05:40.880 very complicated scoring system. That means that even if you're 0:05:40.960 --> 0:05:44.080 very very close to winning, if you suddenly lose that point, 0:05:44.720 --> 0:05:46.080 you go to the end of the line and you 0:05:46.160 --> 0:05:49.359 might not get another chance to play again. And the 0:05:49.480 --> 0:05:53.280 betting system, it's a paramutual odds what exactly to that man. 0:05:53.360 --> 0:05:56.560 So paramutual means that you're betting against the other players, 0:05:56.600 --> 0:05:58.880 and it's not me betting against the house. If I 0:05:58.920 --> 0:06:01.080 was betting against the house, the odds of me winning 0:06:01.120 --> 0:06:03.920 are very small. That's why the house is usually big. 0:06:04.279 --> 0:06:07.680 But in a para mutual system, what happens is the money, 0:06:07.839 --> 0:06:10.120 all the money that is bet in the accompetition is 0:06:11.160 --> 0:06:14.119 thrown into a pool. The house skims off a fee 0:06:14.240 --> 0:06:18.640 something and the rest is divided among the winners. So 0:06:18.680 --> 0:06:21.280 in order to you know, to have a successful betting system, 0:06:21.480 --> 0:06:24.039 you have to be better than the other players. So 0:06:24.160 --> 0:06:26.200 the other betters like basically there are a bunch of 0:06:26.320 --> 0:06:28.960 dumb people like me and my family who used to 0:06:29.000 --> 0:06:30.760 go there from time to time and bet and we 0:06:30.800 --> 0:06:34.039 didn't know anything. And so theoretically, if you're really smart 0:06:34.120 --> 0:06:37.200 and studied, we're the fish that you could take advantag. 0:06:37.200 --> 0:06:38.840 That was the attraction. I mean again, it's it's a 0:06:39.080 --> 0:06:41.920 very exciting sport. It's probably a hard sport to know 0:06:42.600 --> 0:06:45.200 the players very well. You know, I don't think most fans. 0:06:45.560 --> 0:06:48.279 Most fans are not that intense, but they go once 0:06:48.320 --> 0:06:50.480 a year, twice a year. And before we get to 0:06:50.880 --> 0:06:54.160 your sort of rigorous approach. A story in your book. 0:06:54.279 --> 0:06:57.480 You won the first ever bet you placed on Highlights, right, So, 0:06:57.480 --> 0:06:59.479 so the reason we got really hooked on this was 0:06:59.520 --> 0:07:02.160 that when and our parents drove down the floor and 0:07:02.240 --> 0:07:04.840 let us go to highlight one night, they also let 0:07:04.920 --> 0:07:06.919 us make one bet. They gave us two dollars and 0:07:06.960 --> 0:07:11.040 they said, you make one bet, and we followed the 0:07:11.080 --> 0:07:14.080 bet that was listed in the local tow sheet, knowing nothing, 0:07:14.640 --> 0:07:18.119 and astonishingly, it was a trifecta combination of first place 0:07:18.120 --> 0:07:21.400 and show that astonishingly one. And so we won, you know, 0:07:21.400 --> 0:07:24.160 a hundred four dollars. And this was an amazing amount 0:07:24.200 --> 0:07:26.800 of money to a bunch of kids back in the seventies. 0:07:27.200 --> 0:07:30.960 And uh, that's what I was, probably about twelve or 0:07:31.000 --> 0:07:33.480 so at the time. Turning two hundred dollars into a 0:07:33.520 --> 0:07:39.240 hundreds exciting any time, but but when you're a kid 0:07:39.280 --> 0:07:42.160 in the seventies, it must be absolutely Uh. I could 0:07:42.200 --> 0:07:44.000 see how you would then get hooked for life on 0:07:44.000 --> 0:07:47.560 the game. Um, all right, let's fast forward a little bit, 0:07:47.960 --> 0:07:51.880 and so you're a professor at a Stony Brook. Talk 0:07:52.000 --> 0:07:56.280 us about how you started on your path to systematizing 0:07:56.640 --> 0:07:59.720 a gambling system. For highlight and what you did. So 0:08:00.200 --> 0:08:03.080 when you look at the scoring system again, a highlight 0:08:03.120 --> 0:08:07.200 game is played in discreete points player one place player two. 0:08:07.280 --> 0:08:09.560 One of them wins, the other goes to the end 0:08:09.560 --> 0:08:13.080 of the line. You could imagine simulating the result of 0:08:13.120 --> 0:08:16.920 a highlight match by flipping a coin for every particular 0:08:17.000 --> 0:08:20.680 point player one. If player one is maybe better than 0:08:20.720 --> 0:08:23.240 player two, maybe you'd say it as a six chance 0:08:23.280 --> 0:08:26.600 of winning the first point. And if you could figure 0:08:26.600 --> 0:08:29.640 out the odds that one player is going that every 0:08:29.640 --> 0:08:33.240 player has against every other player in that they might 0:08:33.360 --> 0:08:37.400 encounter in a match, you can build a simulation to 0:08:37.720 --> 0:08:41.280 use random numbers to play through and simulate each match. 0:08:41.720 --> 0:08:44.080 So is it like it sounds like a series of 0:08:44.160 --> 0:08:48.000 tree charts almost right? You assigned probabilities for each outcome 0:08:48.040 --> 0:08:50.680 and then you have them sort of branching across all 0:08:50.720 --> 0:08:53.440 the possible outcomes. Right, so you you could view this 0:08:53.480 --> 0:08:56.000 as a tree process. It's a it is a branching process. 0:08:56.000 --> 0:08:59.640 It's a tree process where at every point in the 0:08:59.640 --> 0:09:02.040 tree is every note in the tree is basically two 0:09:02.120 --> 0:09:04.760 players playing each other with a certain score and a 0:09:04.880 --> 0:09:09.200 certain status of players in the queue to come. Then 0:09:09.520 --> 0:09:11.640 depending upon who wins it, you go to a different 0:09:11.640 --> 0:09:14.040 state in the process, and this process ends when you 0:09:14.120 --> 0:09:16.679 have identified who comes in for a second and third. 0:09:16.960 --> 0:09:19.680 So this part sounds is where it seems to really 0:09:19.720 --> 0:09:22.640 diverge from say horse racing, where you just have one event, 0:09:22.920 --> 0:09:25.880 there's one, two, and three, not really all these different 0:09:25.880 --> 0:09:29.040 permutations and sequences. So horse racing is not at the 0:09:29.080 --> 0:09:31.200 scret event kind of a game. This is maybe a 0:09:31.200 --> 0:09:35.079 little bit more akin to I would say, um baseball 0:09:35.120 --> 0:09:37.920 than football. Baseball is a bunch of the scret events. 0:09:37.960 --> 0:09:41.840 There's pitches and things happen. In basketball, things are very continuous. 0:09:41.840 --> 0:09:44.840 In horse racing, things seem continue with So when you 0:09:44.880 --> 0:09:48.079 started developing the system, when are we talking about how 0:09:48.120 --> 0:09:50.360 long ago is this? This is something we started in 0:09:50.400 --> 0:09:52.439 the I would say early nineties. If you have to 0:09:52.520 --> 0:09:54.520 get back there, it's probably the story about in the 0:09:54.559 --> 0:09:57.920 early nineties. Okay, so you broke highlight down into this 0:09:58.080 --> 0:10:01.600 series of discrete events, then what's next in terms of 0:10:01.640 --> 0:10:04.240 the creation of your gambling system. So again, once you 0:10:04.320 --> 0:10:07.319 have the ability to view this as this tree process 0:10:07.440 --> 0:10:10.520 or as this um you can simulate one game, you 0:10:10.559 --> 0:10:13.560 can now simulate a million games or or or more 0:10:14.000 --> 0:10:17.640 and see what this probability distribution is of outcomes, and 0:10:17.720 --> 0:10:22.800 you can start to look at for every combination of first, second, 0:10:22.800 --> 0:10:26.560 and third, how often did it come in? And from 0:10:26.559 --> 0:10:30.040 this that gives you some insight into what things you 0:10:30.080 --> 0:10:33.040 should bet on. But things get a little bit more complicated. First, 0:10:33.120 --> 0:10:36.400 you have to accurately model how good the players are, 0:10:36.679 --> 0:10:38.520 so you have a good guess as to how often 0:10:38.880 --> 0:10:41.920 player one is gonna be player two. And more that 0:10:42.040 --> 0:10:44.040 you have to get a build a model of how 0:10:44.080 --> 0:10:46.360 the public is going to bet um. Again, it's a 0:10:46.360 --> 0:10:51.080 paramutual system. I'm betting against the public. If everyone else 0:10:51.080 --> 0:10:54.120 in the public was someone who programmed the computer and 0:10:54.160 --> 0:10:55.960 did the analysis the way I did, I would have 0:10:56.040 --> 0:11:00.240 no advantage. Are you looking for almost pricing descrepants? Ease 0:11:00.280 --> 0:11:02.560 between where you think the outcome of the game is 0:11:02.559 --> 0:11:05.199 going to come and where people are actually betting exactly 0:11:05.320 --> 0:11:08.400 so that that that again there is a a based 0:11:08.400 --> 0:11:13.680 on our simulation, basically an underlying real probability distribution which 0:11:13.679 --> 0:11:15.960 would in some sense and for a price as to 0:11:16.000 --> 0:11:19.199 what would be a fair return for a two dollar 0:11:19.320 --> 0:11:22.400 bet on that outcome. And then you know we look 0:11:22.440 --> 0:11:25.160 for pricing disreferences, so let's talk about them. Are there 0:11:25.200 --> 0:11:28.640 some persistent biases that you learned about in how the 0:11:28.679 --> 0:11:31.240 public bets? Like you know, I'm the public is silly 0:11:31.360 --> 0:11:35.079 enough to mostly bet on by powerball tickets. People are irrational, 0:11:35.360 --> 0:11:37.600 So what kind of irrationality is did you see that 0:11:37.640 --> 0:11:40.440 you can take advantage of? What one interesting property is 0:11:40.480 --> 0:11:44.880 that since it's very very hard for all the players 0:11:45.000 --> 0:11:47.920 with high uniform numbers to do very well. If you 0:11:47.920 --> 0:11:52.160 could imagine players six, seven, and eight, in order for six, seven, 0:11:52.200 --> 0:11:54.760 and eight to do well well, six has to beat 0:11:55.280 --> 0:11:57.439 seven in order to do well, but that puts seven 0:11:57.440 --> 0:11:59.880 at the end of the line, and seven has to 0:12:00.040 --> 0:12:01.840 eat eight, but that puts eight at the end of 0:12:01.880 --> 0:12:04.120 the line. It's very very hard for there to be 0:12:04.200 --> 0:12:08.000 combinations where all the big numbers come in and essentially 0:12:08.040 --> 0:12:11.120 almost impossible. And yet you would always see people betting 0:12:11.120 --> 0:12:12.800 on this because they didn't know that that way, they 0:12:12.800 --> 0:12:15.880 were betting on essentially an outcome that essentially couldn't happen, 0:12:15.960 --> 0:12:18.640 and the odds must look pretty juicy for those characters. 0:12:18.640 --> 0:12:20.440 For those players, I mean, and it's up to the 0:12:20.440 --> 0:12:22.760 payoff would be good if they want. Because there's only 0:12:22.760 --> 0:12:25.280 one person betting on that during the course of any match, 0:12:25.600 --> 0:12:29.280 but it's never gonna happen. So so so here's our 0:12:29.320 --> 0:12:31.880 system would look for these disrepencies. The other thing that 0:12:31.920 --> 0:12:34.560 we would look for that's sort of related to the 0:12:34.640 --> 0:12:36.720 kind of models people build for trading, is we'd have 0:12:36.760 --> 0:12:41.080 to look at what our impact on the betting pool is. So, um, 0:12:41.120 --> 0:12:44.800 if we bet on something and we win, it doesn't 0:12:44.800 --> 0:12:46.480 pay for us to bet a lot of money on 0:12:46.520 --> 0:12:49.880 that outcome because we're just dividing the pool among all 0:12:49.880 --> 0:12:53.240 the winning tickets, and so every subsequent ticket we would 0:12:53.280 --> 0:12:56.600 buy would would have a lower and lower expected value. 0:12:56.760 --> 0:12:59.880 All right, So you have the mathematics, all the trees, 0:13:00.160 --> 0:13:03.479 you have the nature of how people bet, so the payoffs. 0:13:03.640 --> 0:13:06.640 You also have this calculation about how your own bets 0:13:06.679 --> 0:13:09.600 might affect the results. Now let's talk about putting it 0:13:09.679 --> 0:13:12.760 into practice. What did you do then? So the question 0:13:12.800 --> 0:13:14.800 now is how do we actually go bet on this thing? 0:13:14.800 --> 0:13:17.679 We we couldn't have somebody stationed at the HIGHLFE front 0:13:17.760 --> 0:13:20.640 on every day making our bets the front and we 0:13:20.640 --> 0:13:22.600 were interested in betting in was in Connecticut, I was 0:13:22.640 --> 0:13:24.720 in New York. This was not you know, I'm not 0:13:24.760 --> 0:13:28.160 that crazy, but it turned out that Connecticut did have 0:13:28.200 --> 0:13:31.000 an O TB and off track betting operation where they 0:13:31.040 --> 0:13:33.640 had a phone system where you could dial it in 0:13:33.960 --> 0:13:37.040 and dial in your bets. And so we programmed the 0:13:37.040 --> 0:13:40.760 computer computers back then had these things called modems for 0:13:41.200 --> 0:13:44.080 the kids listening, which I remember that I remember the 0:13:44.120 --> 0:13:46.760 A O L dial up modem, these dial up modems, 0:13:46.760 --> 0:13:49.199 and so you could in some sense, therefore, you had 0:13:49.200 --> 0:13:51.760 a device that you could program that could make phone 0:13:51.800 --> 0:13:55.520 calls and could likewise push buttons in some sense push 0:13:55.600 --> 0:13:58.400 buttons on phones. And so we built a system that 0:13:58.440 --> 0:14:02.320 would um take our bets and convert that into the 0:14:02.400 --> 0:14:05.640 dial tone instructions that would be necessary to place this 0:14:05.720 --> 0:14:08.760 bet at the Connecticut oft TACH betting operation. And so 0:14:08.800 --> 0:14:12.280 we built essentially a complete, you know, programmed trading system 0:14:12.360 --> 0:14:15.680 in highlight. Every day, it would identify go over the web, 0:14:15.720 --> 0:14:20.080 identify what we're the game matches, and who was playing. 0:14:20.360 --> 0:14:23.120 It would simulate each one a million times. It would 0:14:23.160 --> 0:14:26.040 determine the most profitable betting outcomes, and then it would 0:14:26.040 --> 0:14:29.440 phone it into O T B attempt to implement our trade. 0:14:29.520 --> 0:14:33.560 This is amazing. This sounds like algorithmically driven high frequency 0:14:33.920 --> 0:14:36.640 highlight trading essentially. What I love about that is that 0:14:36.680 --> 0:14:39.480 you're exactly right, except there's this very old school part 0:14:39.560 --> 0:14:43.320 because there's high frequency algorithmic highlight trading except for the 0:14:43.400 --> 0:14:46.160 very last part, and it involves dial tones going through 0:14:46.160 --> 0:14:49.680 a phone tree, and so it's this incredibly modern seeming 0:14:49.720 --> 0:14:52.480 idea and then this very old school actually process of 0:14:52.520 --> 0:14:55.160 placing the trades at the very end. I can't say 0:14:55.200 --> 0:14:57.280 that the you know, the the We wrote a book 0:14:57.280 --> 0:15:01.480 about this book called Calculated Bets and market for it 0:15:01.600 --> 0:15:04.200 ended up being not high live fans because it's not 0:15:04.240 --> 0:15:06.440 actually a big universe, but but if they're out to be, 0:15:06.680 --> 0:15:09.000 yet a lot of play in people who were doing 0:15:09.320 --> 0:15:12.760 trading and building these program trading operations, because it is 0:15:13.240 --> 0:15:15.640 essentially the same idea that people are using in the 0:15:15.680 --> 0:15:19.160 same technologies, and it kind of explains basically how these 0:15:19.200 --> 0:15:21.520 things work. I want to just step back and ask 0:15:21.600 --> 0:15:24.440 the dumb question, how would you do? What's an algorithm? 0:15:24.600 --> 0:15:27.480 We hear it all the time, algorithmic trading, and people 0:15:27.520 --> 0:15:29.920 have some idea that it means computers in math, But 0:15:30.000 --> 0:15:32.600 what does this actually mean for someone in plain English? 0:15:32.640 --> 0:15:34.760 As it as it is said in in the world 0:15:34.800 --> 0:15:38.440 of algorithmic trading. It is typically a it's really I guess, 0:15:38.440 --> 0:15:42.680 a programmed procedure from making decisions so that uh, you know, 0:15:42.720 --> 0:15:46.160 there is a decision in a program trading system. There 0:15:46.200 --> 0:15:49.240 has to be somebody making a decision to buy or 0:15:49.280 --> 0:15:52.040 sell this particular stock at this particular time. So a 0:15:52.080 --> 0:15:54.160 set of rules basically, so it can be sort of 0:15:54.200 --> 0:15:56.640 a set of rules as usually some level of input. 0:15:56.840 --> 0:15:59.080 I would say, it could be a simulation, it could 0:15:59.120 --> 0:16:02.480 be sets of rules. It's some kind of a procedure 0:16:02.920 --> 0:16:05.400 that that the sides that this bet or this series 0:16:05.440 --> 0:16:09.160 of bets are profitable, and goes and executes them without 0:16:09.320 --> 0:16:13.320 human involvement. And now you set up the trading system, 0:16:13.400 --> 0:16:16.960 they're dialed in. How much money did you make? Well, 0:16:16.840 --> 0:16:20.400 we may we made okay by percentage wise. Again, recognize 0:16:20.440 --> 0:16:23.120 that the betting pool and highlight is very small. So 0:16:23.200 --> 0:16:26.000 that I told you that a you know, making too 0:16:26.080 --> 0:16:29.080 many bets on any particular match would would rapidly saturate 0:16:29.120 --> 0:16:31.080 the pool and none of the bets would be profitable. 0:16:31.680 --> 0:16:33.480 But but over the course of our trade, we made 0:16:33.480 --> 0:16:38.680 over return on our investment. It was sizeable enough that 0:16:38.720 --> 0:16:41.800 it got to be a little scary to UM run 0:16:41.840 --> 0:16:44.760 it on university research machines, and so we eventually ended. 0:16:44.840 --> 0:16:47.360 It wasn't so much that I have deep regrets about 0:16:47.360 --> 0:16:50.840 turning the system off. What time frame we we had 0:16:50.840 --> 0:16:54.760 it running for about a six month period. Six months, Yes, 0:16:55.000 --> 0:16:57.320 would you would you have gotten to the point if 0:16:57.320 --> 0:16:59.560 you had continued with it where you would just be 0:16:59.640 --> 0:17:03.080 the entire your market UM if we kept um again, 0:17:03.080 --> 0:17:04.960 if if we made our bets bigger and bigger than 0:17:05.000 --> 0:17:07.119 we could have easily become the entire market you know. 0:17:07.440 --> 0:17:09.600 In fact, one of the one of the things that 0:17:09.720 --> 0:17:11.960 was key to our systems again we our system bet 0:17:12.000 --> 0:17:16.640 on Trifecta's combinations of wind, Place and show, which are 0:17:17.040 --> 0:17:20.560 rare events and but but really didn't make it payoff 0:17:20.640 --> 0:17:22.440 was they had a special type of bet called to 0:17:22.480 --> 0:17:26.040 Trifecta box where we could buy a particular set of 0:17:26.080 --> 0:17:30.720 all combinations of three numbers UM cheaper than we could 0:17:30.760 --> 0:17:32.840 the corresponding tickets. And our goal was really to have 0:17:32.880 --> 0:17:35.600 as little impact on the pool as possible, and that 0:17:35.640 --> 0:17:37.920 was really what was necessary. It really wasn't a big 0:17:37.960 --> 0:17:40.560 margin here. If someone wants to do this, I mean, 0:17:40.600 --> 0:17:42.840 I know that it's harder, but what are the key 0:17:42.920 --> 0:17:46.240 areas of mathematics to study. So again I am a 0:17:46.240 --> 0:17:49.679 computer scientist, and so in this case there was um, 0:17:49.720 --> 0:17:52.840 you know, to understand things about Monte Carlo simulations. Again 0:17:52.840 --> 0:17:55.560 we were talking about this tree process. Um, you could 0:17:55.640 --> 0:17:59.280 view this as building a tree that you exhaustively analyze 0:17:59.280 --> 0:18:01.480 where you could simul late at using something called Monte 0:18:01.560 --> 0:18:04.639 Carlo simulation, where you really did use random numbers to 0:18:04.760 --> 0:18:07.960 describe the path down the trade. So knowing computer science 0:18:08.040 --> 0:18:11.600 is a good thing. Um, you know, knowing something about statistics. 0:18:11.920 --> 0:18:14.240 You know, data science is a new field called data science, 0:18:14.240 --> 0:18:16.520 which is the kind of area where my my lab works. 0:18:16.760 --> 0:18:18.960 And uh then this kind of field that this is 0:18:18.960 --> 0:18:20.680 the kind of stuff that I think is good learning 0:18:20.680 --> 0:18:23.600 how to build models, this kind of thing. So you 0:18:23.640 --> 0:18:26.239 mentioned you came out of this book Calculated Beds, and 0:18:26.280 --> 0:18:28.800 it wasn't a huge hit among high life fans because 0:18:28.840 --> 0:18:31.359 there aren't that many high life fans, but it got 0:18:31.400 --> 0:18:34.240 a lot of followers among people who play the market, 0:18:34.359 --> 0:18:37.280 people in banks. What are some of the key lessons 0:18:37.320 --> 0:18:39.239 in terms of what you did and how else they 0:18:39.280 --> 0:18:42.359 apply to someone wanting to play the markets and setting 0:18:42.400 --> 0:18:45.320 up a trading system. So the first thing that I 0:18:45.359 --> 0:18:50.560 would say is that that markets are in general relatively efficient, 0:18:50.720 --> 0:18:54.040 even in high Lie where we had you know, these 0:18:54.119 --> 0:18:57.320 crazy you know, the people who were watching embedding workers only, 0:18:57.359 --> 0:18:59.760 these people who went once a year and didn't know anything, 0:18:59.840 --> 0:19:02.600 the pools of dumb money. I was surprised how hard 0:19:02.640 --> 0:19:04.880 it was for us to build a system that actually 0:19:04.960 --> 0:19:07.960 did um, did have a positive return. I thought it 0:19:08.000 --> 0:19:10.479 was gonna be a lot easier than that than it 0:19:10.480 --> 0:19:14.000 turned out to be. And that's that's probably a lesson 0:19:14.040 --> 0:19:17.440 that most markets are are more efficient than you would think. 0:19:17.680 --> 0:19:19.520 You know, even in horse ray. They've been done studies 0:19:19.560 --> 0:19:23.000 in horse racing, and the markets there are relatively efficient. 0:19:23.040 --> 0:19:26.119 You know, the fact is that there's a large um 0:19:26.280 --> 0:19:30.800 transaction cost essentially of the house keeps, and that's a 0:19:30.880 --> 0:19:34.600 large transaction cost, and that's white people lose. So certain 0:19:34.640 --> 0:19:36.840 things are models are harder to build than than you 0:19:36.840 --> 0:19:39.720 would think. That's I guess one lesson here. The other 0:19:39.920 --> 0:19:42.720 is that if you're careful and you're you're thinking hard enough, 0:19:42.760 --> 0:19:44.560 and you beat on it and maybe there's something there. 0:19:44.960 --> 0:19:49.040 Does the lesson of not saturating the market apply to 0:19:49.280 --> 0:19:52.639 broader financial markets because obviously we hear a lot today 0:19:53.200 --> 0:19:57.400 about um high frequency trading, algorithmically driven trading, and it's 0:19:57.440 --> 0:20:00.200 impact on markets. What do you think, Yeah, so it's 0:20:00.200 --> 0:20:03.240 certainly the case that in any market, you if you 0:20:03.320 --> 0:20:07.600 bet enough, you're eventually betting against yourself. And uh, the 0:20:07.600 --> 0:20:10.800 the advantage of the financial markets that they're generally large 0:20:10.920 --> 0:20:13.760 enough that you can put in a tremendous amount of 0:20:13.760 --> 0:20:16.680 capital and play before it, you know, you really start 0:20:16.720 --> 0:20:19.720 betting against yourself. But again, you know many in many 0:20:19.800 --> 0:20:24.639 hedge funds, in some sense there's they will occasionally occasionally 0:20:24.640 --> 0:20:26.920 a hedge funds will return capital if they can't think 0:20:26.920 --> 0:20:29.960 they can invest it efficiently enough. And that's basically because 0:20:30.000 --> 0:20:32.360 of these saturation effects. But what about if you get 0:20:32.359 --> 0:20:36.880 a market that becomes dominated by algorithmic trading and they 0:20:36.920 --> 0:20:39.280 all kind of feed on each other. Does that end 0:20:39.320 --> 0:20:43.040 up having the same effect. It's an interesting question. Um, 0:20:43.080 --> 0:20:47.440 if everybody was doing the exact same thing in a market, 0:20:47.840 --> 0:20:51.159 then there wouldn't be an interesting market going on. And 0:20:51.200 --> 0:20:54.720 so the question of whether whether algorithmic trading is going 0:20:54.760 --> 0:20:58.600 to eventually get into a world where you know, sometimes 0:20:58.800 --> 0:21:01.240 people are betting against them where nothing is happening, It 0:21:01.320 --> 0:21:04.639 depends upon the traders doing different things. I guess markets 0:21:04.640 --> 0:21:08.160 would only get into trouble if all the different traders 0:21:08.160 --> 0:21:10.320 were doing the exact same things. That's I guess when 0:21:10.320 --> 0:21:14.040 you get into bubbles and when you get into into crashes, 0:21:14.080 --> 0:21:17.840