WEBVTT - How Jim Simons Built a Machine That Beat the Market 0:00:17.560 --> 0:00:21.000 Pushkin too quick. No, it's perfect Pushkin stuff. 0:00:21.400 --> 0:00:37.959 You got it, Robert Smith. Yeah, tell me about the 0:00:38.000 --> 0:00:38.440 Cold War. 0:00:39.120 --> 0:00:41.599 The Cold War between the United States and the Soviet 0:00:41.680 --> 0:00:45.479 Union thrust us into this technological world that we live 0:00:45.479 --> 0:00:49.960 in today. We know about how it brought us the Internet, satellites, 0:00:50.120 --> 0:00:53.960 silicon chips, but the Cold War also brought us the 0:00:54.000 --> 0:00:59.560 modern financial market system. Stocks, bonds, all traded with high 0:00:59.600 --> 0:01:04.919 frequency computers, right algorithms. This quant world of Wall Street 0:01:05.240 --> 0:01:07.719 came from the Cold War, and the one man most 0:01:07.720 --> 0:01:12.560 responsible for this financial breakthrough was a genius mathematician who 0:01:12.600 --> 0:01:16.120 grew disillusion with the US government and decided to create 0:01:16.160 --> 0:01:21.760 the most efficient money making machine ever built great setup 0:01:21.920 --> 0:01:25.039 for his own profit. By the way, so it's the 0:01:25.040 --> 0:01:30.440 early nineteen sixties and Jim Simons is studying math. Like 0:01:30.440 --> 0:01:31.960 a lot of people in the United States. There was 0:01:32.000 --> 0:01:35.040 this real push after Sputnik for people to go into 0:01:35.040 --> 0:01:40.240 math and science. Jim Simons graduates from MIT from UC Berkeley, 0:01:40.880 --> 0:01:44.360 and he decides that he doesn't really want to do 0:01:44.440 --> 0:01:47.080 mathematics of the physical world. He wants to do the 0:01:47.200 --> 0:01:53.560 highest form theoretical mathematics. His thesis topic on the transitivity 0:01:53.560 --> 0:01:56.280 of holonomi systems or hulminy. 0:01:56.400 --> 0:01:57.800 Do you know what that word means? 0:01:58.080 --> 0:02:02.960 I do not, eh no, But why do you read 0:02:03.000 --> 0:02:05.440 the first Why do you read the first two sentences 0:02:05.480 --> 0:02:07.040 of the first theorem so we can get a feel 0:02:07.080 --> 0:02:07.280 for it. 0:02:07.600 --> 0:02:09.040 Yeah, put it here, you've sent it to me. It 0:02:09.040 --> 0:02:15.080 says theorem one. Let M be a c to the infinity. 0:02:15.600 --> 0:02:20.000 I guess manifold with an affine connection. Let R denote 0:02:20.040 --> 0:02:22.640 the curvative tensor of the connection. 0:02:23.680 --> 0:02:24.560 I don't know what it means. 0:02:24.639 --> 0:02:26.800 I know you could even have us something to the infinity. 0:02:27.080 --> 0:02:31.440 But I'll tell you who loved this, the US government 0:02:31.639 --> 0:02:33.760 and all the people who worked for the US government. 0:02:34.080 --> 0:02:36.000 And he was recruited to work at something called the 0:02:36.360 --> 0:02:39.079 Institute for Defense Analyzes in Princeton. 0:02:39.280 --> 0:02:42.240 Why don't they just call it the CIA math. 0:02:42.360 --> 0:02:46.680 Shop front at Tecton? Is it that technically independent? It 0:02:46.760 --> 0:02:49.560 is a think tank that helps the US government during 0:02:49.600 --> 0:02:52.800 the Cold War. They do things with you know, weapons 0:02:52.919 --> 0:02:56.720 targeting and evaluation, but they also do code breaking, and 0:02:56.760 --> 0:03:00.919 that's where they put Jim Simons, the mathematician. There's anything 0:03:00.960 --> 0:03:03.040 wrong with that? So if you think about it, right, 0:03:03.360 --> 0:03:05.440 because we know this is a show about finance eventually, 0:03:06.160 --> 0:03:08.560 but think about what he's doing in order to break 0:03:08.720 --> 0:03:13.280 Soviet codes. Jim Simons is looking looking at a sea 0:03:13.320 --> 0:03:17.800 of random signals, seemingly random data, but he knows that 0:03:17.840 --> 0:03:21.200 there's a communication inside. He knows there's a pattern, there's 0:03:21.240 --> 0:03:25.760 a signal in the noise, and he's using sophisticated mathematics, 0:03:26.000 --> 0:03:30.600 algorithms and early computers, big computers to churn through this 0:03:30.720 --> 0:03:34.200 data and try and figure out what the Soviets are saying. 0:03:34.280 --> 0:03:36.600 What is the data telling us, what is going to 0:03:36.640 --> 0:03:38.600 happen in the world exactly? 0:03:38.720 --> 0:03:41.800 Now, By all accounts, Jim Simons is pretty good at 0:03:41.800 --> 0:03:45.240 this hard to tell. It's all secret, right, And in 0:03:45.280 --> 0:03:47.480 another world he might have been one of those mathematicians 0:03:47.520 --> 0:03:50.720 who stayed on at the NSA, had an illustrious career, 0:03:51.120 --> 0:03:52.640 and we never would have heard of him because it 0:03:52.680 --> 0:03:56.280 all would have been secret. But instead he did something amazing. 0:03:56.760 --> 0:03:59.240 So this is during the Vietnam War, the early days 0:03:59.240 --> 0:04:02.320 of the Vietnam War, right, and the head of the 0:04:02.360 --> 0:04:07.040 think tank the IDA was publicly defending US military action 0:04:07.160 --> 0:04:10.160 in Vietnam, saying we are winning the war. And Jim 0:04:10.240 --> 0:04:14.040 Simons is young mathematician. The Times in his late twenties. Right, 0:04:14.360 --> 0:04:18.279 he thinks it's total crap, and he writes a public 0:04:18.400 --> 0:04:22.039 letter to the New York Times magazine published there that says, 0:04:22.360 --> 0:04:26.120 the war in Vietnam is a waste of time and 0:04:26.159 --> 0:04:29.240 money and human lives. Quote, it would make us stronger 0:04:29.279 --> 0:04:32.920 to construct decent transportation on our East coast than it 0:04:32.960 --> 0:04:35.799 would be to destroy all the bridges in Vietnam. Huh, 0:04:36.000 --> 0:04:39.520 I know true today? Right? Needless to say, he writes this, 0:04:39.800 --> 0:04:42.880 criticizing his boss, he is fired from the thing. Sure, 0:04:43.320 --> 0:04:47.240 reasonably so, one might say, But Jim Simon's already had 0:04:47.360 --> 0:04:50.360 this idea that would change finance. What if you could 0:04:50.440 --> 0:04:54.200 use these same systems, computers, algorithms, code breaking, right to 0:04:54.279 --> 0:04:57.479 look for patterns in stocks and bonds. He would go 0:04:57.520 --> 0:05:01.080 on to create a firm called Renaissance Technologies, with one 0:05:01.200 --> 0:05:05.640 fund in particular, called the Medallion Fund. Over the span 0:05:06.400 --> 0:05:09.960 of thirty years or so, the Medallion Fund would return 0:05:10.160 --> 0:05:15.120 just a staggering amount of money. It's estimated that its 0:05:15.320 --> 0:05:20.360 trading profits were more than one hundred billion dollars. Just 0:05:20.480 --> 0:05:23.680 you know, the stock market averages about eleven percent gain 0:05:23.800 --> 0:05:26.960 a year, right, seven percent after inflation? Maybe? 0:05:27.120 --> 0:05:27.400 Right? 0:05:28.040 --> 0:05:33.240 Jim Simons was producing returns of sixty six percent a. 0:05:33.240 --> 0:05:35.040 Year for thirty years. 0:05:36.360 --> 0:05:39.200 That sounds wrong to me. 0:05:39.360 --> 0:05:42.000 That sounds like somebody did a made a math there, right, 0:05:42.080 --> 0:05:45.839 Like if you think of Bernie made off Right, the 0:05:45.880 --> 0:05:48.800 famous Ponzi schemer, he was returning like what ten to 0:05:48.839 --> 0:05:50.479 twelve percent a year and. 0:05:50.960 --> 0:05:52.680 That was his dream. That was fake, and. 0:05:52.640 --> 0:05:56.680 That was he was cheating to do that, right, sixty 0:05:56.760 --> 0:05:59.120 six percent a year is so for thirty years? Like sure, 0:05:59.120 --> 0:06:00.320 if somebody could do it for a couple of years, 0:06:00.320 --> 0:06:03.360 they get lucky. Like, are you sure that is right? 0:06:03.839 --> 0:06:06.719 This is as far as we can tell an accurate 0:06:06.800 --> 0:06:12.440 number between nineteen eighty eight and twenty twenty one. Can't 0:06:12.640 --> 0:06:15.760 believe it. Double your money in sixteen months and. 0:06:15.720 --> 0:06:17.840 Then double it again and again and again and again 0:06:17.880 --> 0:06:20.920 and again. Infinite money machine keeps doing it at that 0:06:20.960 --> 0:06:24.480 magnitude is extraordinary. 0:06:25.520 --> 0:06:29.560 This is Business History, a show about the history of business. 0:06:29.720 --> 0:06:32.440 I'm Jacob Goldstein, I'm Robert Smith. Today on the show, 0:06:32.480 --> 0:06:35.960 the final part of our investing series, we featured way 0:06:36.040 --> 0:06:40.200 back more than one hundred years ago, the speculator Jesse Livermore, 0:06:40.440 --> 0:06:44.320 the great investor Warren Buffett, love him and now the 0:06:44.400 --> 0:06:47.520 quant Jim Simons, he was part of a revolution in 0:06:47.600 --> 0:06:51.400 finance to put math and numbers before all this intuition 0:06:51.560 --> 0:06:54.159 and business savvy stuff that Wall Street was all about, 0:06:54.160 --> 0:06:58.320 before this right, he built this system that created enormous wealth. 0:06:58.839 --> 0:07:03.080 And the amazing thing is that no one really knows 0:07:03.440 --> 0:07:07.120 how the system works, how they're getting that return. Even 0:07:07.160 --> 0:07:10.440 Jim Simons, the man who built it, doesn't quite know 0:07:10.840 --> 0:07:13.320 what it's doing. When you see a picture of Jim 0:07:13.360 --> 0:07:16.120 Simon as you'd never be like, oh, that's a fancy 0:07:16.240 --> 0:07:20.720 Wall Street hotshot, like he looks like a mathematician. Continued 0:07:20.760 --> 0:07:23.440 to look like a mathematician until the day he died. 0:07:23.520 --> 0:07:27.760 He had the tweed coat, the thick glasses, the wispy hair, 0:07:27.960 --> 0:07:32.080 right penny loafers. He was always smoking. At some points 0:07:32.080 --> 0:07:35.600 he would smoke like three packs of Merit cigarettes a day. 0:07:35.920 --> 0:07:37.760 A machine for smoking cigarettes. 0:07:37.960 --> 0:07:40.320 It's a secret machine. And after that letter to the 0:07:40.400 --> 0:07:44.720 editor about Vietnam, Jim Simons lived a very secretive life. 0:07:44.760 --> 0:07:47.640 He didn't really talk to the press talk about himself. Eventually, 0:07:47.640 --> 0:07:50.040 when he created this money machine, he didn't tell people 0:07:50.080 --> 0:07:52.240 how he did it. He didn't want people to visit him, 0:07:52.600 --> 0:07:54.400 you know, at his investment firm. 0:07:54.920 --> 0:07:57.200 He just wanted to smoke cigarettes and make money. 0:07:57.320 --> 0:08:01.960 Well, and I'm all out of cigarettes. So a lot 0:08:02.000 --> 0:08:05.040 of the story that we're going to tell today comes 0:08:05.040 --> 0:08:08.000 from a great investigative reporter, Greg Zuckerman of The Wall 0:08:08.000 --> 0:08:10.800 Street Journal. He wrote a biography of Simon's called The 0:08:10.840 --> 0:08:14.960 Man Who Solved the Market. So, after getting fired, Jim 0:08:15.000 --> 0:08:19.080 Simons goes to work in academia the State University of 0:08:19.320 --> 0:08:23.960 New York at Stony Brooks Stonybrook University, and by all accounts, 0:08:24.000 --> 0:08:26.400 he was a very good mathematician, but he was an 0:08:26.480 --> 0:08:31.440 even better manager of mathematicians. As head of the department 0:08:31.440 --> 0:08:34.280 there at Stonybrook University, he would travel the country and 0:08:34.360 --> 0:08:37.240 look for young rising stars and convince him to come 0:08:37.280 --> 0:08:41.280 to this math program on Long Island that turned out 0:08:41.319 --> 0:08:43.200 to be one of the best in the country, right, 0:08:43.240 --> 0:08:46.599 and he would bring them all together. Spoiler, some of 0:08:46.640 --> 0:08:50.200 those mathematicians he would later lure to his fund. This 0:08:50.320 --> 0:08:53.160 became one of his real big strengths. So he does 0:08:53.200 --> 0:08:56.520 this for a while. Nineteen seventy eight, Simons is forty 0:08:56.720 --> 0:08:59.719 years old. He's kind of old for a mathematician, right, 0:08:59.840 --> 0:09:03.560 He decides to leave the university because he already looks 0:09:03.960 --> 0:09:06.040 the part. Everyone thinks maybe he's retired. 0:09:06.760 --> 0:09:09.080 But although cigarettes make him look sixty. 0:09:08.720 --> 0:09:12.880 What exactly right? Although he was not going to retire, 0:09:12.920 --> 0:09:15.880 he was going to finally do this idea that he had. 0:09:15.960 --> 0:09:18.680 He was going to use computers to solve the market. 0:09:19.160 --> 0:09:23.400 And it turns out it is much harder than it looks, right, 0:09:23.559 --> 0:09:27.240 like one idea, you're like. Nowadays we think of how 0:09:27.280 --> 0:09:30.600 powerful computers are all the data we have, but back then, 0:09:31.360 --> 0:09:34.480 just having the idea didn't mean that you could pull 0:09:34.520 --> 0:09:34.880 it off. 0:09:35.160 --> 0:09:39.880 That is a classic business truth. Right, Like ideas are cheap, execution. 0:09:39.520 --> 0:09:43.040 Is hard, and in fact, even for Simon's himself, it 0:09:43.080 --> 0:09:45.560 would take probably more than a decade before he could 0:09:45.600 --> 0:09:48.240 really figure out how to do this. Simons teams up 0:09:48.240 --> 0:09:50.440 with a fellow math genius he had met at the 0:09:50.480 --> 0:09:55.240 Institute for Defense Analyses, Lenny Bomb the Bomb. The Bomb 0:09:55.640 --> 0:09:59.720 Bomb's an expert in something called hidden markoff processes, which 0:09:59.840 --> 0:10:06.040 was this way of using probabilistic outcomes of random processes 0:10:06.200 --> 0:10:09.959 to determine hidden that's that's as much as I have 0:10:10.240 --> 0:10:12.680 I was with you. Well, they we're doing what we 0:10:12.679 --> 0:10:15.000 would call today machine learning. 0:10:15.520 --> 0:10:19.439 Machine learning is like, essentially, you build a system in 0:10:19.480 --> 0:10:21.440 a computer. You feeded a bunch of data, and the 0:10:21.480 --> 0:10:24.400 system sort of builds a map of the relationships in 0:10:24.440 --> 0:10:26.760 that data, and then with new data it can kind 0:10:26.800 --> 0:10:30.400 of interpolate or extrapolate and make guesses about what should 0:10:30.400 --> 0:10:33.520 come next. And of course today we have an exciting, 0:10:33.760 --> 0:10:36.840 maybe misleading, confounding term for machine learning. 0:10:37.360 --> 0:10:39.880 We call it AI exactly right, right, And so they 0:10:39.880 --> 0:10:43.440 start this firm called Monometrics. But as you say, the 0:10:43.520 --> 0:10:46.680 key is not the math. They have math in spades. 0:10:46.760 --> 0:10:49.320 What they don't have is they don't have the data. 0:10:49.760 --> 0:10:54.559 Also a classic modern AI problem. Like every AI founder 0:10:54.600 --> 0:10:58.600 I've interviewed, it always turns into a story about collecting 0:10:58.679 --> 0:11:00.640 the data, building the data set. 0:11:00.760 --> 0:11:03.560 And this is the nineteen seventies, right, So data is 0:11:03.679 --> 0:11:06.880 not at the tip of your fingers on a computer keyboard. 0:11:07.040 --> 0:11:11.920 You had to physically go and find data, which is amazing. Right. 0:11:12.160 --> 0:11:15.280 So they have the closing prices every day for you know, 0:11:15.360 --> 0:11:18.199 commodities and bonds and stocks. This was published in the 0:11:18.200 --> 0:11:20.160 Wall Street chart, so everyone had that data. 0:11:20.160 --> 0:11:21.439 It's like one one. 0:11:21.360 --> 0:11:25.000 Number per day per stock, per commodity. Yeah, but what 0:11:25.040 --> 0:11:27.400 you're trying to do is find correlations. You're trying to 0:11:27.440 --> 0:11:31.480 find things in the real world that impact those prices. 0:11:31.880 --> 0:11:34.400 What's the data set for that? They had to go 0:11:34.880 --> 0:11:38.200 look for it. So for instance, they would buy old 0:11:38.280 --> 0:11:44.600 magnetic tapes from commodities trading exchanges, like big giant computer 0:11:44.679 --> 0:11:46.360 magnetic tapes, and they had to figure out how to 0:11:46.360 --> 0:11:48.760 get the data off of it. They would buy stacks 0:11:48.800 --> 0:11:51.119 of books from the World Bank. 0:11:51.120 --> 0:11:53.800 Huh, like reports that the World Bank, like what is 0:11:53.840 --> 0:11:55.920 what are the reports about what countries were doing and 0:11:56.000 --> 0:11:57.440 exchange rates and things like that. 0:11:57.920 --> 0:12:00.400 They had a stafford go to the Federal Reserve offices 0:12:00.440 --> 0:12:03.800 to like write down interest rate changes over the years. 0:12:04.559 --> 0:12:07.440 Because once you have that, then the computer can start 0:12:07.440 --> 0:12:09.360 looking for pattern right. And to be clear, they're not 0:12:09.480 --> 0:12:14.200 like studying these reports and these day they're just in 0:12:14.640 --> 0:12:18.520 shoveling it in, right, And this is still very basic 0:12:18.600 --> 0:12:22.200 machine learning, and the computer at the time keeps making 0:12:22.280 --> 0:12:26.760 mistakes that they didn't really understand. So, for instance, once 0:12:26.840 --> 0:12:33.440 the computer developed a taste for potatoes, main potatoes, so 0:12:33.600 --> 0:12:36.760 Zuckerman tells his story in his book, the system kept 0:12:36.800 --> 0:12:41.240 buying main potato futures in the state of Maine. 0:12:41.720 --> 0:12:44.880 Potato potatoes big harvest year next year or whatever. 0:12:45.120 --> 0:12:49.200 Yes, okay, until two thirds of the company's money was 0:12:49.240 --> 0:12:52.080 in potatoes. They were all in potatoes. And they got 0:12:52.080 --> 0:12:54.240 a call from the regulators, the cft A. 0:12:54.280 --> 0:12:57.800 CFPP commod Futures Trading Commission. 0:12:57.440 --> 0:13:01.160 Right yeah, saying whoa, who are you guys, Like, what 0:13:01.200 --> 0:13:03.560 are you doing over there? You have almost cornered the 0:13:03.600 --> 0:13:07.520 market on potatoes. You have to sell. And they ended 0:13:07.600 --> 0:13:11.800 up losing money on the trade because because blown out 0:13:11.840 --> 0:13:15.320 on potatoes, they had stopped the computer or whatever the 0:13:15.360 --> 0:13:18.559 computer's plan was. But you know, this was just one 0:13:18.800 --> 0:13:23.520 small weird thing. Simon and Baum were really kind of 0:13:23.679 --> 0:13:26.640 nervous about this whole thing. They had taken investors money, 0:13:26.800 --> 0:13:30.600 they didn't really know if their system worked, and as 0:13:30.640 --> 0:13:33.560 the story gets told, they start to like second guess 0:13:33.600 --> 0:13:36.079 the computer and themselves, and they start to think, well, 0:13:36.360 --> 0:13:38.960 I have this intuition that gold's gonna go up because 0:13:39.000 --> 0:13:42.480 of the geopolitical situation, and they'd make some money on that, 0:13:42.600 --> 0:13:44.760 and then they'd lose some money on that, and so 0:13:45.520 --> 0:13:49.040 by doubting their own system, it just wasn't really working. 0:13:49.120 --> 0:13:52.559 And as they say, like it was super stressful because 0:13:52.640 --> 0:13:55.520 like at that point They're just Wall Street investors, right 0:13:55.559 --> 0:13:58.360 with the big computer trying to buy more potatoes, and 0:13:58.480 --> 0:14:01.400 the and the man won't let them buy potatoes. It 0:14:01.440 --> 0:14:05.320 wasn't really the mathematical based system that Jim Simons had 0:14:05.360 --> 0:14:05.800 dreamed of. 0:14:06.240 --> 0:14:08.760 I mean, in a way, it makes sense, right even 0:14:08.760 --> 0:14:12.800 for them, because what they're trying to do is so 0:14:13.080 --> 0:14:17.800 contrary to human nature. Right, Like human nature, especially if 0:14:17.800 --> 0:14:20.720 you are smart, like Jim Simons clearly is is. Well, 0:14:20.840 --> 0:14:23.920 I'm smart. I can look at the market and see 0:14:23.920 --> 0:14:26.720 what's gonna happen. I can understand what drives the price 0:14:26.760 --> 0:14:28.840 of gold and where it's going, and I'm just gonna 0:14:28.880 --> 0:14:31.200 make a bet. I'm going to bet on you know 0:14:31.280 --> 0:14:34.520 my understanding. Right, that is human nature, And what they 0:14:34.560 --> 0:14:37.760 fundamentally want to do is, in a way take themselves 0:14:37.800 --> 0:14:40.360 out of the equation. Right, say, I'm going to build 0:14:40.400 --> 0:14:43.160 a computer and then I'm not gonna use my gut. 0:14:43.160 --> 0:14:45.080 I'm just gonna do what the computer says. And so 0:14:45.120 --> 0:14:47.680 in a way, it's not surprising that even Jim Simons 0:14:48.040 --> 0:14:52.760 can't quite fully commit to taking himself out of the equation. 0:14:52.920 --> 0:14:55.320 You turn on the TV news there's an oil embargo, 0:14:55.360 --> 0:14:57.480 and you think, oh, does my computer know about this 0:14:57.640 --> 0:15:00.120 now is able to do it like I should sell 0:15:00.240 --> 0:15:04.200 or I should buy? Right, So monometrics doesn't really work out, 0:15:04.520 --> 0:15:08.520 and in nineteen eighty two, Jim Simons starts a new company, 0:15:08.560 --> 0:15:13.080 Renaissance Technologies and Renaissance Technologies. He thought, well, it's going 0:15:13.160 --> 0:15:15.440 to be more of a tech investing firm, you know, 0:15:15.640 --> 0:15:17.640 VC kind of thing, which would be great in the 0:15:17.720 --> 0:15:21.960 nineteen eighties, right, But he kept the computer trading idea going, 0:15:22.160 --> 0:15:25.720 eventually folding it into something that would become this amazing fund. 0:15:25.800 --> 0:15:29.680 We talked about the Medallion Fund. And the first thing 0:15:30.280 --> 0:15:34.600 that Jim Simon's was just really good at was recruiting 0:15:34.640 --> 0:15:37.520 the right people to come into Wall Street, which was 0:15:37.600 --> 0:15:39.880 unexpected at the time. Now we're like, oh, you're a 0:15:39.880 --> 0:15:44.000 mathematics major at Princeton University, good luck at Golden Sacks, right, 0:15:44.640 --> 0:15:47.800 But at the time this was somewhat unusual. And he 0:15:47.880 --> 0:15:55.880 starts hiring his fellow mathematicians, quantum physicists, linguists, number theorist, astronomers, sure, 0:15:56.000 --> 0:15:58.040 which if you think about like astronomers looking at all 0:15:58.040 --> 0:15:59.960 this data in the sky and has to find out, 0:16:00.120 --> 0:16:02.520 you know, where a black hole is. These are all 0:16:02.600 --> 0:16:05.320 people who can see the signal in the noise, and 0:16:05.360 --> 0:16:08.440 he's bringing them out to Long Island, right, And he 0:16:08.520 --> 0:16:11.760 had this insight that I think only a mathematician could have, 0:16:12.000 --> 0:16:16.960 which is math geeks are really competitive. You know, we 0:16:17.040 --> 0:16:20.600 may not think of that, but they really want to 0:16:20.640 --> 0:16:23.440 win and they want to prove these like impossible theorems. 0:16:24.040 --> 0:16:28.280 And mathematicians, you know, frankly, if you haven't made major 0:16:28.320 --> 0:16:30.520 awards by the age of thirty or thirty five, they 0:16:30.600 --> 0:16:32.360 kind of feel like their career is a little bit over, 0:16:32.760 --> 0:16:35.720 so they are open to moving to an investment firm. 0:16:36.160 --> 0:16:39.840 It reminds me a little bit of you know, Paul Graham, 0:16:39.920 --> 0:16:42.640 he was the founder or a founder of y Combinator, 0:16:42.680 --> 0:16:47.440 the startup incubator. He wrote this whatever essay blog post 0:16:47.480 --> 0:16:52.160 called Fierce Nerds a few years ago. Fierce Nerds this phrase, 0:16:52.240 --> 0:16:56.080 and his point was like, traditionally in sort of culture, 0:16:56.120 --> 0:16:59.520 the nerd was portrayed as kind of deferential. You know, 0:16:59.600 --> 0:17:01.880 the jack is maybe the alpha, and the nerd is 0:17:01.960 --> 0:17:05.399 kind of subservient or whatever. But there is this type 0:17:05.440 --> 0:17:08.760 that he identified, the fierce nerd that's like the alpha nerd, 0:17:09.000 --> 0:17:11.400 the nerd that really wants to win. To show the 0:17:11.440 --> 0:17:14.679 world that they are smarter, better and like it's a 0:17:14.679 --> 0:17:17.760 classic founder type, right, like the tech founder. But now 0:17:17.840 --> 0:17:21.320 also yeah, now it has become but it's also this 0:17:21.400 --> 0:17:25.320 type of kind of competitive mathematician who who Simons had 0:17:25.400 --> 0:17:26.560 spotted and recruited. 0:17:27.600 --> 0:17:31.000 The second major thing I think Jim Simons was doing 0:17:31.720 --> 0:17:36.120 is this relentless focus on data, on hoovering up more 0:17:36.160 --> 0:17:39.959 and more data. They got really good at finding, you know, 0:17:40.080 --> 0:17:43.479 intra day prices, little tick data they call it stocks 0:17:43.480 --> 0:17:45.560 going up going down all throughout the day, not just 0:17:45.600 --> 0:17:48.359 the closing price. You know. They were looking at newspapers 0:17:48.400 --> 0:17:51.040 to get the news items in there, all nacs, old 0:17:51.080 --> 0:17:51.880 punch cards. 0:17:52.080 --> 0:17:54.119 And when you say looking at you mean in putting 0:17:54.160 --> 0:17:57.080 into the system, right, that's just more data, more data. 0:17:56.920 --> 0:18:01.720 Yes, and more importantly, they were good at cleaning up 0:18:01.800 --> 0:18:04.480 the data. And this is something you don't really think about, 0:18:04.520 --> 0:18:07.320 but there's mistakes all over the place, there's mission gaps 0:18:07.400 --> 0:18:10.800 in all forms of data. Figuring that out and figuring 0:18:10.800 --> 0:18:14.359 out what to put in its place, called cleaning the 0:18:14.440 --> 0:18:19.040 data is a mathematical problem, and they were very good 0:18:19.119 --> 0:18:19.480 at it. 0:18:19.800 --> 0:18:23.040 But so simple, such a simple idea, and yet obviously 0:18:23.119 --> 0:18:23.840 so powerful. 0:18:24.040 --> 0:18:27.480 You know, I've downloaded data sets that were like two 0:18:27.560 --> 0:18:30.119 hundred thousand items when I was in business school. And 0:18:30.160 --> 0:18:33.040 when you have two hundred thousand items, you can't find 0:18:33.080 --> 0:18:35.800 a missing cell in there or someone who's like put 0:18:35.840 --> 0:18:38.680 a wrong number in there. Now imagine millions and millions 0:18:38.760 --> 0:18:43.240 and millions of pieces of data. It's very hard to 0:18:43.359 --> 0:18:46.320 not make something that will stop a computer dead in 0:18:46.359 --> 0:18:49.760 its tracks. Right. And so even with all of this, 0:18:50.440 --> 0:18:53.800 it did take them years and years and years to 0:18:53.880 --> 0:18:56.240 really start making money with this system. You know, they 0:18:56.280 --> 0:18:58.480 would place these big bats on a move in the 0:18:58.480 --> 0:19:01.159 stock market. Sometimes it would work, sometimes it wouldn't, and 0:19:01.240 --> 0:19:05.280 they couldn't figure out why. And they finally brought in somebody, 0:19:05.680 --> 0:19:09.200 a game theorist that helped make it work. His name 0:19:09.280 --> 0:19:14.800 was Elwin Burlecamp, and he essentially said, you know, the 0:19:14.920 --> 0:19:19.240 problem here is that you're acting too much like an 0:19:19.320 --> 0:19:23.560 investment fund. To make the quant thing work. You need 0:19:23.600 --> 0:19:26.719 to start acting more like a casino. 0:19:29.000 --> 0:19:52.399 Coins falling out of the slot machine after the break. 0:19:58.080 --> 0:20:00.080 That is the end of the break. Now we're going 0:20:00.160 --> 0:20:02.560 to talk about game theory and why it's a good 0:20:02.560 --> 0:20:05.600 tax like a casino. If you are Jim Simons in Renaissance. 0:20:06.080 --> 0:20:09.320 The year is nineteen eighty eight. Jim Simon's quant investing 0:20:09.440 --> 0:20:11.840 fund that started out as Monometrics has changed its name 0:20:11.880 --> 0:20:16.119 to Medallion because of all the mathematical prizes his employees 0:20:16.200 --> 0:20:19.080 have won, including himself. And they brought in this expert 0:20:19.160 --> 0:20:23.000 in game theory. Elwyn Burlecamp and Burlacamp ran in the 0:20:23.000 --> 0:20:27.959 same circles as a famous investor and gambler named Ed Thorpe, 0:20:27.960 --> 0:20:28.680 who you've met. 0:20:28.800 --> 0:20:32.840 I interviewed him. Yeah, he's a super interesting guy. Know 0:20:33.040 --> 0:20:35.919 now as as an investor, he made a ton of 0:20:35.920 --> 0:20:38.760 money as an investor, but he earlier in his career 0:20:39.119 --> 0:20:42.040 went to Vegas, learned how to count cards, wrote a 0:20:42.040 --> 0:20:45.119 book called Beat the Dealer did this amazing thing, and 0:20:45.160 --> 0:20:48.000 I think the sixties with Claude Shannon, another super interesting guy, 0:20:48.000 --> 0:20:51.679 where they built this machine to like understand how the 0:20:51.760 --> 0:20:55.119 roulette ball was going to break and like basically beat roulette. 0:20:55.200 --> 0:20:58.400 Like extremely interesting finance math guy. 0:20:59.119 --> 0:21:03.160 Yeah. So Ellen Burlacamp knows Ed Thorpe and is thinking 0:21:03.200 --> 0:21:06.880 about gambling strategies, right, and he looks at the Medallion 0:21:06.960 --> 0:21:11.199 Fund and he says, you're making all these big bets 0:21:11.560 --> 0:21:14.159 on market. It moves right, and you win sometimes and 0:21:14.200 --> 0:21:17.520 you lose sometimes. But the key is, if you're going 0:21:17.560 --> 0:21:22.359 to rely on probabilities, you need a lot of bets. 0:21:22.920 --> 0:21:27.160 Imagine a casino that rolled the roulette wheel once a day, Right, 0:21:27.240 --> 0:21:29.760 they could win money from the customers. They could lose 0:21:29.800 --> 0:21:32.800 a bunch of money, but you wouldn't know it's random 0:21:32.800 --> 0:21:36.680 at that point. Yes, right, but you do ten thousand 0:21:36.960 --> 0:21:41.000 roulette wheel spins, one hundred thousand, a million roulette wheel spins, 0:21:41.560 --> 0:21:44.680 and the casino has an edge, and they will win 0:21:44.800 --> 0:21:46.639 money after a million turns. 0:21:46.720 --> 0:21:48.720 It's the law of large numbers. If the odds are 0:21:48.720 --> 0:21:51.560 in your favor, you want to be making essentially as 0:21:51.560 --> 0:21:52.640 many bets as possible. 0:21:52.720 --> 0:21:55.440 Yeah, and he says, let's do at the medallion fund 0:21:55.440 --> 0:21:58.840 what casinos do. Get a tiny advantage, figure out how 0:21:58.880 --> 0:22:01.359 to win slightly more than fifty percent of the time, 0:22:01.800 --> 0:22:05.160 and then just make a lot of bets. So instead 0:22:05.200 --> 0:22:07.560 of you know, making one bet and seeing how it's 0:22:07.560 --> 0:22:07.880 going to. 0:22:07.800 --> 0:22:10.080 Turn out, how all it on Maine potatoes and see 0:22:10.119 --> 0:22:11.720 it how the harvest comes out next. 0:22:11.600 --> 0:22:14.159 Year, you know you are going to make bets on 0:22:14.200 --> 0:22:17.720 the market the last maybe a couple of days one 0:22:17.840 --> 0:22:20.960 day intra day, this sort of high speed trading that 0:22:21.000 --> 0:22:24.879 we would eventually see in the market. And here's an 0:22:24.960 --> 0:22:27.359 example of like what they started to see. When they 0:22:27.400 --> 0:22:30.440 did this, they needed just tiny little advantages they could 0:22:30.440 --> 0:22:34.600 exploit in a small way. So the computer spotted something 0:22:34.680 --> 0:22:39.680 that was futures traders out in the market. If they 0:22:40.440 --> 0:22:43.600 had a good week, you know, made money on their 0:22:43.640 --> 0:22:46.800 contracts over the week, they would tend to sell at 0:22:46.840 --> 0:22:50.320 the end of the day on Friday and then rebuy 0:22:50.359 --> 0:22:53.359 the contracts on Monday morning. So if they're selling on Friday, 0:22:53.359 --> 0:22:56.160 the price goes down a little bit. If they're buying 0:22:56.160 --> 0:22:59.200 on Monday, the price goes up a little bit. And 0:22:59.960 --> 0:23:03.520 they didn't really know why. Maybe you know, the future 0:23:03.600 --> 0:23:05.959 traders wanted to chill on the weekend, or you know, 0:23:06.000 --> 0:23:08.439 maybe they were worried about world events. Whatever it was. 0:23:09.359 --> 0:23:11.920 The Medallion Fund was like, well, we're just gonna buy 0:23:12.000 --> 0:23:13.600 a bunch of viewers of contracts on Friday, We're going 0:23:13.680 --> 0:23:15.560