Max Dama: I had founded the, the trading club at Berkeley when it was very, very small, I'm, I should be like, you know, the cheerleader for this, but instead now when I go to Berkeley and I see 150 students in the audience, I'm like, there's not 150 seats in this industry. You know? So even

Steve Hsu: Welcome to Manifold. My guest today is Max Dama. He's the co-chair of Headlands Technology, a quant prop trading firm in Chicago. Max, welcome to the podcast.

Max Dama: Happy to be here. Thanks for having me.

Steve Hsu: Thanks for coming on. I became aware of you, I think, 'cause I, I read some stuff you posted over the years online and we set up a Discord server for this podcast and there's now about, I don't know, six or 700 people on that Discord server, all of whom are super interesting. But then when you came on, a bunch of people recognized you and started asking you a bunch of questions. So I thought, oh, I should get this guy on the show. So thanks for coming on.

Max Dama: Maybe, yeah, maybe two people, but yeah. There's not too many people out there in the trading industry.

Steve Hsu: It must be more than that.

But, let's start with your background. So you, you're from Florida, but you went to boarding school in Tennessee, is that right?

Max Dama: Yeah, right. So it's northwest Florida, like the redneck part.

At some point you just kind of run out of educational options. So my older brother, he was smart. I was not really the smart one, I was just kind of like surfing and chilling. And he was in the Duke Tip program. So he took the SAT when he was very young and then he went to this boarding school and they were like, okay, he's really smart, let's get his brother.

So I went there too. And then I actually became smart in that school. So I kind of learned how to study and got interested in artificial intelligence and trading, and then started to think like, okay, maybe I could start to combine these things. But yeah. Yeah, I think that boarding school's a really good experience.

It's called McCallie. It's in Tennessee. It's all boys. If you have a son, I would recommend studying him here. There, I guess your son's a little bit too old.

Steve Hsu: Yeah, my son's already in college, but interestingly, he wants to be a quant trader like you. Mm-hmm.

Max Dama: Yeah, it's very popular nowadays.

Steve Hsu: Yeah. So that's something I wanna ask you about.

So, I think you just said that in high school you were already interested in AI and the relationship between AI and trading.

Max Dama: Yeah, I was doing trading on my own and I was trying to build some video games on my own. I thought I was like the first one to think that I could combine these two things.

So yeah, eventually I went to Berkeley and then I was still kind of on a different track, but I started a blog about trading. I was trying to kind of develop my trading ideas and share them with the world. This was like the peak of the blogosphere when people were reposting each other and meeting each other in person, going to different conferences.

So people would contact me through that and they'd ask me to help program their trading systems and things. So I got some good opportunities to meet individual traders. Like I flew to Korea to program some trading system for one guy, but I didn't know what I was doing. So I kind of feel bad for him. I don't think he minded that much either, which is some karaoke.

Steve Hsu: So you are already interested in the markets and trading it sounds like, even from high school days.

And is that a common thing among traders? My impression nowadays is that it's kind of a high prestige thing to do if you're kind of a quant oriented person in college. But, is it your experience that a lot of the best traders are people who have been interested in it like kind of like, since they were kids?

Max Dama: Yeah, I mean there's different cycles. So the industry is so young that it's hard to characterize the whole industry that way. But yeah, I think the people that survived in the industry for many years, were generally really passionate about the markets themselves. I think nowadays it's changed a lot.

So nowadays a lot of students are just like, okay, that job pays a lot, so I want to get into it. I'm going to practice all these brain teasers, which I kind of see as a little bit of a negative indicator. So I think it's better if people are just interested in the problems themselves because it is a struggle. Yeah, it's not just like some easy career path.

Steve Hsu: Those are points I want to revisit in just a minute, but I want to focus a little bit on your development as a young person. So you, you went to college at Berkeley pretty far from, you know, the, the south. How did you choose that place and what did you want to do in college?

Max Dama: I went to a summer program at Stanford when I was in high school, and I really liked the West coast. And then applied to Stanford. Didn't get in, got into Berkeley, so went there. But I was very happy with it. I mean, Berkeley was great. There were so many Nobel laureates, so you could, every semester I just tried to take every class that sounded interesting to me and I thought could be applied to trading.

So I ended up taking all the different machine learning courses and randomized algorithms and things that you just couldn't really take at other schools 'cause they wouldn't be big enough. So yeah. Then I did a series of internships and trading, which I had just hustled to find 'em, like they weren't posted anywhere.

Like one guy, I had to write a small trading system for him to apply for the internship. And then that team shut down six months after I left. It was pretty tricky back then. And then I interned for one hedge fund in San Francisco, which was led by Astro Teller who went on to become the head of Google X.

That hedge fund didn't really make much money, but a lot of interesting people there. When I was getting near graduation, I was applying to different firms, but it was very hard to find different firms back then. So got rejected from a few and then found headlands through like a series of connections.

And they were just starting up in San Francisco at that time. So I joined Headlands and I had a trading blog when I was in college. When I joined Headlands, I had to kind of shut down that trading blog. So I convert it to PDF and then release it on the internet. And somehow in the following 15 years, like not many people wrote anything too useful about trading.

So still that PDF kind of circulates and every now and then people contact me about it. So it's, it's been interesting to kind of see that thing have its own life.

Steve Hsu: So that, that PDF, which it's quite long, right? Is it like 60 pages long or something or maybe more?

none: Yeah,

Steve Hsu: Yeah, yeah. So it is like a mini guide, an informal kind of intro to trading.

And I've seen it online. I think you know, not many people are gonna sit down and write something that extensive. Right? So

Max Dama: Yeah, I didn't either, like, I mean, had that blog for four years and then the blog content also became basically a syllabus for a student club or student-led class at Berkeley about trading that I taught for four semesters.

So I was just trying to kind of build up a community and meet people, but it was so small back then nobody really knew about trading. So yeah, that, that kind of community didn't really develop much. I think out of my graduating class, maybe there's only three people left, or my graduating class, plus the year older and the year younger, just three people left.

none:

Max Dama: But yeah, that, that PDF, it wasn't written overnight. It was like the syllabus and that's why I had a nice structure. And then when I released it as a PDF, that was just so that it wouldn't get deleted off the internet.

Steve Hsu: But it, I mean, you are unusual that, you know, you had this interest and it sounds like you even structured your undergraduates, you know, the courses you took and, and, and you spent four semesters teaching a class on trading, so you're like.

You're sort of like the OG trader, right? I mean, how, how many, like, okay, there are stories about Ken Griffin trading bonds from his Harvard dorm room or something, but, but pretty unusual, right? It's not, there are not a lot of people like you.

Max Dama: Yeah, I mean, there, there are plenty out there. Like I've met so many people over the years because once you're in the industry, you know so many people.

So like there's a guy, you can find him online Gary Basin. He's active on Twitter, but he runs his own trading firm from his dorm room. So he was way more successful than me, in terms of just the trading system in college. And I mean, there, there are many, many people like that. So I wouldn't say I'm like the OG or like anything special there.

I mean, everybody who had previous firms prior to me did similar things. So yeah, I had set up my own, like AWS server and I was kind of running my strategy. I don't think it had any alpha, I don't think it was making any money. If I let it run longer, it probably would've been a draw down. So.

Steve Hsu: But now I think from your bio, it says, I think you, you ended up earning, did you earn three simultaneous degrees at Cal?

That's pretty impressive, right?

Max Dama: Yeah. I mean, I ended up with four, so, four,

Steve Hsu: Right?

Max Dama: I was just taking every class. So every semester I take like six or seven classes because I was like, wow, you know, this is a huge smorgasbord of interesting ideas. And at Cal you could take graduate classes if you were just like willing to sit in the pack and then ask the professor every week, Hey, can you enroll me?

Can you enroll me? And then once people drop, so yeah, I ended up, like, near the end, I just looked at my course list and I was like, Hey, I think I have the requirements set, satisfy all these majors. Then I went to the registrar and was like, Hey, can you just write these down? And they're like, no, there's only two entries.

I'm like, just write it. Just write it. And I, I actually didn't think they'd let me graduate with that, but they did. So, but anyway, I wouldn't recommend it. I was just kind of confused about what I wanted to take. Okay.

Steve Hsu: Now, is there anything that stood out to you? Because I'm guessing most of these professors might be experts on machine learning or algorithms or, or applied math, but most of them don't really understand the markets.

I would even say most economists don't understand, you know, things like bid ask and price structure or microstructure. Most people don't really understand that. So were you listening to some guy who's like the world's expert on linear algebra and in the back of your head thinking like, oh, I could use that for trading?

Like, what's your thought process during these years?

Max Dama: Yeah, exactly. I was reframing everything in terms of trading problems. There was one professor, John McAuliffe. He was a student of Michael Jordan, the, the be guy and no net guy. Yes. He had come back to Berkeley to found Von, but at that time I didn't know he was founding Von.

He had just been a professor and I was like, oh, he worked at, I think he worked at the Shaw before that. So I tried to take his class and talk to him every day and he wouldn't talk to me. He didn't wanna share his secrets, so, but yeah, I mean, I would just try to think. On my own, like, okay, how can I use this genetic algorithm or how can I use this support factor machine?

You know, there were no neural nets back then as support factor machines or vision learning.

Steve Hsu: Now, I think there's one now, I can't remember his name, but there's one math or stats Prophet Cal at Berkeley, who, I think he was early in Renaissance. I dunno if that rings a bell with you.

Max Dama: Yeah, Berlekamp

Steve Hsu: Berlekamp that's right.

Elwyn Berlekamp. Yeah. Yeah.

Max Dama: Berlekamp. Yeah. I met him once in Berkeley because he had come back to Berkeley to found a new trading firm. I think it was called, like Berkeley trading. It didn't last very long. And when I met him he was, he was very, very old. So we weren't really able to communicate much.

But yeah, it was an interesting place. Like I think I saw David Shaw walking down campus one time and yeah. So a lot of people around there.

Steve Hsu: Incredible. My academic background because I come from physics in most of my, you know, the kids that I was in college with and grad school and wrote papers with. I think at least half of them are in finance now, but very few of them are really hardcore traders.

because I don't think they had that focus that you had on it. They were more like, oh, I want to be a physicist or something. And then later, um, they realized, oh, they'll pay me a lot of money to price derivatives or something. And so to me, the trading mentality, like the idea that you're gonna develop some system that has alpha and backtest it and do all this, it's something I didn't really actually encounter that much of.

Even though I know tons of people who now currently work in finance, I'm curious if you ever heard of, there was an early physicist, maybe named Osborne, who maybe wrote this like a very old book that talks about market microstructure and stuff. I don't know if you've ever seen this book, but that's like the very first academic type study I've seen on topics related to this kind of thing.

Max Dama: Hmm. Yeah, no, I haven't, yeah, I'm surprised because, I mean, I guess there's just been these different phases of the quant kind of moniker where, so there was a person who wrote a book called, like, My Life as a Quant. Yes. And Emanuel Derman. And so I, I kind of fill in these gaps that were before my time. And I think when people were developing blackshaws and all the different options, pricing models, derivative pricing models, or portfolio optimizers, then they had a big need for physicists.

So a lot of physicists came into the industry, but I think nowadays a lot of those problems are, you know, mostly solved. At least there's not a lot of like, very fundamental research, or just brand new models from scratch. You might be doing modifications of existing models, but like in my generation, there weren't many physicists going into, say, trading.

So I'm more like an HFT trader. It's mostly computer science people. And so maybe in your generation it was like physicists pushing investment bankers. In my generation it was, you know, computer scientists pushing out floor traders. You know, there's the famous picture of the Stanford, Connecticut, UBS trading floor with thousands and thousands of desks.

And then eight years later it's totally empty. So I feel like, yeah, whenever a new technology comes along and then the skill set that you need to, you know, build things with new technology changes, then you see these shifts. But I wouldn't necessarily say that physicists are highly represented in, um, like HFT space, maybe more in SAT or longer frequencies, which maybe we can talk about later.

Steve Hsu: Yeah, I think you're right. I mean, I think it's more, much more CS type people. Now at the time when a lot of my friends were going into finance, um, leaving physics and they ended up doing stuff like options, pricing theory and stuff like that. There was this mysterious group of people out on Long Island, the Renaissance guys, and nobody knew what they were up to.

And I'm curious, when you were coming up, were they still a kind of mysterious, iconic firm for you? Or how, how did you think about them at all?

Max Dama: Yeah, definitely. Yeah, they're still mysterious. I think they're, yeah, one of the most mysterious firms out there. So when I was coming up, I guess the biggest trading firm was Getco.

So I was trying to meet people there, and I, I had some contacts there, so I was trying to organize events with them at Berkeley, but we ended up having to bus down to Stanford. but yeah, they were kind of like doing 50% of the market volume at that time. So at least on the HFT space, they were the big ones.

But yeah, I think Renaissance has always been a little bit longer term, but nobody, nobody really knows, so.

Steve Hsu: Yeah. And do you find, like for the kind of trading that you do, do you end up in a situation where there's only a certain amount of capital you can deploy? Fruitfully and you end up, it ends up being basically just the firm's own money after a while, or, or do you guys still take tons of LP money from outside?

Max Dama: Oh, no, it's very capital constrained. Yeah. I don't think there are any major HFT firms that have outside money. So, I mean, basically what you see is like if, if somebody is trading and then their sharp ratio goes above like a three, then they will buy out the outside investors. and then because basically if your short ratio is three, then maybe the returns are 30%.

So that means every year you're basically getting 30% profit. So you can start to buy out the investors pretty quickly.

Steve Hsu: Right.

Max Dama: Assuming there's not infinite capacity, which there's basically never infinite capacity. So yeah. For, for HFT, if the returns on capital are higher, then you can buy them out even quicker.

So sometimes a new startup will have some outside investors for like a year or two and they'll have some special structure and then they'll buy it out eventually.

Steve Hsu: Got it. So when you first started, Were you confident? Were you, did you say, Hey man, I'm gonna go in there and I know I'm gonna figure something out.

I'm gonna have some serious alpha, or were, were you, did you have a worry in the back of your mind, like, uh, some guys are gonna back me and I'm gonna start training, but maybe it's not gonna go well for me? Like, what, what were you thinking?

Max Dama: Mm, yeah, I didn't really know what to expect going into the industry. So, I mean, the way it works is things are a lot more collaborative. It's not like you have your own strategy and you have a backer. So everybody's working together. We're building a large software system together. Um, I think I definitely had a lot of imposter syndrome for many years and probably I was kind of an imposter.

Like I just didn't know what was going on and I had to ask people for help with Bel Floss and things like that. So I think it was also healthy to be modest and humble. But no, I mean, I always liked building the trade system, so I always just liked showing up every day and having that opportunity.

Steve Hsu: How much of the barrier to getting into this is the infrastructure, like the software infrastructure, it sounds like a huge amount of energy and talent goes into that, and not all firms, I would guess, can actually build good infrastructure.

Max Dama: Yeah, for sure. Yeah, I think that's basically what it is. It's building infrastructure

Steve Hsu: And so as a guy who can write code and help you build that infrastructure, is that guy as valuable or more valuable than somebody who, you know, is really good at statistical analysis and has lots of ideas?

Like what, how, what's the kind of trade off between those different talent types?

Max Dama: Yeah, I think I saw Elon just posted like, there's no more researcher title at Grok or Tesla or SpaceX. I don't know which company he was talking about. It's just all engineers. So yeah, I think it's kind of similar in the sense that like, if, if you're a researcher, you have to implement your ideas.

And at some point they kind of outgrowth the existing system. And so when you generalize the existing system, you need to do some, some work to make it actually function in practice. Or you encounter some, you know, barrier like cap theorem where it's distributed and there's just no way to satisfy all the constraints.

So if you have some imaginary idea that you want to try, then maybe it's actually not even feasible. So, yeah, I always think the people that go the farthest in the industry are kind of able to think at, at the systems level and all different levels of the stack. Yeah, it's basically a technology business.

So it's similar to, you know, Google, Facebook, something like that. And that's usually the type of people that we'll hire or that's the, those opportunities that people don't like trading, then they'll go to Google or Facebook.

Steve Hsu: So you, you see a lot of movement back and forth between, or you did in the past between big tech and, and trading firms?

Max Dama: Yeah. Yeah. I mean, I think basically like. Historically, maybe there was like a, you know, a trading industry and it kind of looked like a trading industry. There's guys in the pit, or there are people trading on screens and they're making kind of discretionary decisions. then over the years, every year, like the amount of the percent of volume trading in the pit would decrease like about 5%.

So on each market it could be agriculture or energy markets, bond markets. And some were faster, some were slower, but basically it'd be like roughly 5%. And then also, you know, COVID accelerated it. So that was maybe like 10% for the ones that hadn't transitioned fully. But once things tipped to like 80% electronic, then basically all the input to the system is fully digital.

And then you can just, you know, think of everything digitally and you can just build models. So like, you know, Google, basically a digitalized library and then all the, you know, people there can make a lot of money because it's like a very scalable business where they can. Index all the information with just a team of like a hundred people.

And I think it's similar with HFT, where you have a small group of people that as long as all the information going into the system's digital and all the things you're sending out to the market are just digital, you don't have any sales people or anything, then it becomes Yeah, very, very scalable and it's just building a software system.

Steve Hsu: Got it. So it seems like to me, like for a young person just getting outta college like 10 years ago, that the dream job was kind of big tech or something, and now it seems like talking to my son that, you know, Jane Street or something like that is like what every kid want, well, every kid in a certain category, uh, wants to do.

So maybe you can talk me through what happened there? Like how did this become, how did, how did all these smart kids at all these different schools become aware of just a few firms where they wanna work at?

Max Dama: Yeah, I might have a contrary opinion there compared to other people in the industry. So like, I think there's a bit of a lagging effect with, with recruiting where

People are always reacting to the offers that they heard in the past or they posted online in the past. And so, yeah, trading firms were giving really large offers and they were hiring a lot. And the industry is also very, very cyclical. So whenever you know Donald Trump gets elected, then you know, people in the trading industry are like, okay, there's more volatility coming.

So, you know, the profit in the industry is roughly proportional to the volume times of volatility because like you kind of are participating with the rest of the market. So every price, movement and every other trade, you kind of participate with it. So whenever there's something like, like a war or like Trump, then kind of the, the overall industry profit kind of increases for a while, but it's cyclical, so it's not gonna last forever.

And like having worked 15 years, I've seen a lot of different cycles. So yeah, I think basically there was just a big boom like a few years ago because of a confluence of events. And then a lot of people got interested in the industry. And because I had founded the trading club at Berkeley when it was very, very small, I'm, I should be like, you know, the cheerleader for this. But instead now when I go to Berkeley and I see 150 students in the audience, I'm like, there's not 150 seats in this industry. You know? So even maybe there's only 10 big firms, and each of those firms only has, you know, like maybe 50 people who are kind of like building a lot of the models. There's obviously more people than that, but like the jobs that people really want, it's a smaller number.

So yeah, it's not a big industry and I think probably the amount of interest by the kind of undergraduates right now is a little bit overdone.

Steve Hsu: I thought I saw a number saying like, if you look at like, you know, market making as an activity and just you just sort of estimate what's up for grabs in this whole system.

That it could only be like maybe $30 billion a year. Am I hallucinating?

Max Dama: Yeah. I think a few years ago I would've said it's lower, but I think, yeah, maybe like recently it seems a little higher. I don't know if that's inflation or what. Yeah, something like that. I think that's a pretty good ballpark number.

none: Right.

Max Dama: So, so yeah, if you compare it to other industries, like the shampoo industry, you know, it's, it's smaller than, you know, something like that.

Steve Hsu: Right. So I think maybe the, maybe I saw that in your defense of HFT, like saying that people are making a lot of money and there are, there are people going into it, but it's not distorting all these other industries 'cause it's just too small.

Max Dama: Yeah. It's weird how like, in one sense it gets a lot of attention and the other sense it gets a lot less attention. So yeah. I'm not sure why it has that perception.

Steve Hsu: Yeah. I mean, I'm guilty myself because I am one of these guys 'cause so many of my friends went into finance, not specifically the type of trading that you do, but as we said. a lot of talent, like a lot of people who in a previous era or if just The macro dynamics were a little bit better for academic research jobs. They would've stayed in some kind of frontier science, but they ended up in finance. Over many years I've written like, wow, my, I'm not sure that these people are being, I mean, they're, they are, their jobs require a lot of talent, but I'm not sure that they're doing the best thing possible for the overall economy or for society.

So I'm definitely guilty of that, although maybe not focused that much on HFT.

Max Dama: Yeah, like I said, I can only think of, I was trying to think before this call of who I know from Berkeley that's still in trading. And I mean, I could think of like John Zhu, he's the head of trading at Optiver and yeah, maybe another guy who's a partner at IMC, Raj Patel.

But most people stayed in the industry for maybe like two or three years and then left.

Steve Hsu: Well, that's super interesting.

Max Dama: Like I said earlier, once people digitalized trading, then it was amenable to, you know, people with the skillset.

So I think now a lot of other industries are being digitalized. So, you know, Google did it for information and obviously Google's revenue is orders of magnitude more than trading. So everybody wants to work there and there's, you know, 20,000 employees or something. Facebook also, you know, digitalized friendship.

So now people with our skillset can go there. And I think, yeah, way more people are going to those areas, especially from my cohort. And then with AI it will kind of digitalize a lot of the real world. So, you know, through robotics or just through language or vision. So I think those will be the big areas for kind of like people that see themselves as intellectuals or like to operate in information space rather than something more like people or selling or things

Steve Hsu: like that.

So the, the kids, you went to college with it, so you said some, like a typical modal time to last in the industry is only a few years.

Max Dama: Yeah, definitely. Yeah. And that's why now when, you know, when I talk to people, I'm like, if, if you're not really interested in the industry, like probably you should find something else.

Like if you're not at least trying to build your own trading system nowadays, you know, it's popular to build some small marketing systems when you're in college. Yeah. If you're not building some trading system on your own, then there's gonna be really hard years where, you know, there's high expectation and maybe the market's down or something.

And so if, if people are not passionate about it and I don't think it's going to work out too well.

Steve Hsu: Now there are slots in there for people who just like to, you know, maybe they're not that interested in trading per se, but they, they just like building systems and, you know, they're software developers.

Are you, including those guys in this category or just the people who actually like run strategies?

Max Dama: Yeah, I mean, there's a lot of different jobs, right? We have accountants, we have,

none: you know,

Max Dama: HR, like, so there's all the normal jobs. And also, yeah, certainly if, if people like other things, then there's still a lot of places for them, you know, if you just like to code some low LA system, like obviously it's in high demand and things like that.

But yeah, I just think if you want to kind of be running a trading firm or like always be thinking about the next step, then you kind of need to be interested in the problem domain to solve.

Steve Hsu: And so for the kind of stuff you do, are you guys almost, I mean, typically market neutral, like you're not predicting directional stuff the way like a macro guy would, right?

You're kind of operating without any assumptions like that.

Max Dama: So I think there's a lot of things in trading where once you work in the industry, you kind of look at things on a very, very micro scale. So you look at things on a packet by packet basis. And then any notion of simultaneously kind of disappears. So market neutral kind of assumes simultaneously, but when you place a trade, like you're sending a network packet over the wire and if you send one to buy, one to sell, like they reach at different times and so you can try to time it. Do people try to time it in ht? Not usually like I think in the past people would really try to remain market neutral, but then that was when they had a lot of edge and they could afford to cross the spread twice just to kind of reduce the risk.

And then over time, like the edge disappeared. And so people generally just tried to kind of remain hedged, roughly, like kind of just maybe if you're trading frequently enough kind of by accident, you get some central limit theorem and things cancel out. Or maybe after five minutes you eventually find a hedge, you know, rough hedge.

But it's not, it's not like one-to-one hedging or things like that.

Steve Hsu: Got it. So the way you describe that is really interesting. You're sending out packets over the network and buying a cell and you're sort of just trying to get an edge on that tiny mechanistic event, right?

You're not, you're not thinking about summing over like hours and hours of this stuff, but just, just that microscopic thing. Am I understanding that right?

Max Dama: Yeah. Yeah. I would say if you're thinking about things on that level, then you're high frequency, and if you're thinking about things on a higher level, then you're, you know, some other frequency, like your stat is mid frequency.

People have different words for these things

Steve Hsu: Got it. Yeah.

Max Dama: I would say that's kind of like the distinctive thing of just like whether you're thinking about individual price levels and individual packets.

Steve Hsu: So at that level where you're operating the skillset, you know, is it more like

Steve Hsu: thinking about algorithms in computer science or is it more like thinking about pure math? What's the number one thing you would look for in a hire, in an interview?

Max Dama: Yeah.

There's not like one thing that you can learn in academia that's going to map directly to trading.

So I think, yeah, you need to be interested in trading. I think there's, in terms of what you can learn in academia that's useful for trading, like so, yeah, I had a couple majors, so like in terms of the math major, there's linear algebra and numerical analysis. It's kind of like how do you actually implement things with double flow point precision and avoid error accumulation things and algorithms.

For math, that's probably it. You know, like it helps take a proof class so you can just kind of think logically. For statistics like the trading domain, it's not really anything like IID or driven by any statistical process. It's just that these tools happen to work. So it helps to have that toolbox of things like linear aggression, trees, neural networks, and you kind of just try them with some kind of very weird parameters that you wouldn't normally find in statistical applications, like a lot of regularization or things like that.

But yeah, in terms of statistics, if you just take maybe like a couple of classes there too, to kind of get an intuition for probability distributions and uncertainty, raising about uncertainty and some kind of machine learning toolbox and then computer science.

I think everything in computer science is useful because you're just encountering all these problems on a dayday basis.

And it's an interesting situation where you actually do have to optimize for everything. There is no premature optimization, like you have to

none: make

Max Dama: it fast, you have to make it fit in memory. Both for kind of the inference step, you know, the production trading step and also the research process where you're iterating over historical data.

So yeah, everything in computer science is useful. Implementing a programming language. Berkeley has a very infamous class that used to be taught by this guy Hilfinger. I gravitated toward that class. you know, imp implementing an operating system because for trading you have to bypass everything in the operating system and kind of like reimplement it yourself network protocols, like for trading, you always implement your own network protocols to try to step up, strip off all the fights in the header and avoid, like rate limiting things.

And then obviously the lower level ones on just data structures and software engineering.

Steve Hsu: So, you know what you just said, it makes me think you understand this area literally from the ground up, from the bits from the electrons up. Is that true for everybody in your industry or are you special in that way?

Max Dama: Yeah, that's true for everybody. Yeah, not special. And I think there's even some people that know it a lot better than me. So there's a lot of tricks that people will use to kind. And you can read about 'em online every now and then. So especially if you read, what's a good one? So, US Exchange, UEX Exchange is the largest features exchange in Germany .

And they will publish some white papers about the architecture of their matching engine and they'll explain why they're changing their architecture. And it's because somebody is sending a packet to them and the header is not quite fully completed because they want to lock up a spot in the switch and they'll just complete it later.

So that gives them a few nanoseconds to speed up, like things like that. So I've never had to deal with things like that, but there are definitely people that are even lower level than me.

Steve Hsu: I think there was some old case maybe involving Renaissance and two Russian guys who had been trained in physics and there was some lawsuit and if you looked at the documents they were.

Claiming they were actually told to do things, which, you know, could be considered hack, like network hacking on, on maybe the, the market making server or something. Anyway, back in the day, this was like a big deal, I think. But it sounds a lot like what you're describing now.

Max Dama: Yeah. They're, they're always these things that pop up and they're not that huge in the industry.

Like, they're not that significant. Like most of it's just having a really good system that does everything well. But it is fun to talk about these little things. Like, I think at one point maybe Singapore had banned microwave lines, but somebody had set up their own microwave and a tower and then the competitor found out about it through kind of like interviewing people that firm and they put aluminum foil outside the tower to block it

none: because

Max Dama: like them, they can't complain to the authorities because what they were doing was also illegal.

So just some crazy things like that. There was a movie about some of these things called Hummingbird Project. I think it had Toma Hayek.

Steve Hsu: Oh, I think I saw that. Yes. Big

Max Dama: actors. It was not a commercial success, I don't think, but. If you're in the industry, it's kind of fun.

Steve Hsu: Okay, tell me if this is wrong, but it seems to me that you, you need to be above some level of ability, you know, to master all these different things in the stack that you're talking about, but it seems like interest and obsessiveness is probably as or more important.

And so this idea that like, oh, you need to be an IMO gold medalist to do this kind of work. Like, is that just misleading? I mean how important is that kind of a genius to this kind of job?

Max Dama: I mean, I think those people can probably do a lot of things well, so I think for me, I definitely didn't have that. So I was just interested in the problem and I think that was enough for me.

none: Yeah.

Max Dama: But yeah, I mean, I have worked with people that are much smarter than me and I, I think there's been a different phase in my career where I was like, okay, there is a 10 x engineer, there's not a 10 x engineer.

Like I thought before I went to Berkeley, I thought. Okay. There are definitely super geniuses out there. And then I went to Berkeley and I did four majors and I'm like, I'm not very smart. 'cause I even did math competitions and I failed in 'em. Right. And I had an older brother who was way smarter than me, so I was like, okay.

Then I was at Berkeley. I'm like, okay, actually there's not any super genius people. And then I worked at Headlands and you know, then met some super genius people. So I think there's been different phases. I think maybe at Berkeley I just had been interacting with the super genius people. They were probably there, but I just wasn't interacting with them because I was maybe interested in trading or something.

So

Steve Hsu: Yeah, maybe the super genius people were all trying to learn topology or something, or string theory maybe.

Max Dama: Yeah, I think trading is a very abstract thing because yeah, at some level you are just thinking about these packets and other participants and things and all the data you're receiving in your system is just packets on a wire and everything you send out is packets on a wire.

So everything you do in between is just kind of all very abstract and theoretical. So it definitely helps just to be able to think theoretically.

Steve Hsu: So you're interviewing a candidate and are you giving them brain teasers or are you not giving them brain teasers?

Max Dama: Me personally, I'm not because I remember when I was graduating and I was so bad at them, and then in my everyday work, it doesn't feel very applicable.

I'm not like trying to do mental math on the fly. Everything is completely automated and I can take as much time as I want and I can just come in on the weekend and try to grind through a problem. So I'm not, I think there are still quite a few companies doing that. And I mean, I think as an IQ test it might make sense for me.

Part of the aversion is also, I just think they're a little high variance, so if you ask the same question to like equal people on two different days, you might get different answers.

Steve Hsu: I agree with you. I think brain teasers are high variants and so. I've never relied on that in interviewing people.

Usually when I interview someone, I might give them something to think about and then I wanna see how they think about it. But it's usually not some very well posed cut and dried thing where they have to figure out one trick or two tricks, and then they can solve it. I just never had a lot of faith in that.

Max Dama: Yeah. I think over the years I came to appreciate having a very diverse set of interviews for all the candidates because some people always look good according to certain interview criteria, so you just need a bunch of different angles on things. So I don't really try to tell people to do it a certain way,

Steve Hsu: Do we have a generation of kids who like studying brain teasers in order to get through the, you know, first three Jane Street, you know, filters or something?

Is that, that's a thing, right?

Max Dama: Yeah. Yeah. And I, that's the thing I regret of all writing my PDF, is that I had those interview questions in there. And I, I don't think people should be spending their time on that. I think they should be spending their time trying to master all the material that they can learn in school, in academia, because there is a lot of really, really good material that has been synthesized.

I mean, just to have the way Berkeley taught, computer science was teaching you every single layer of the stack. And the fact that they could build that curriculum within just like 30 years while they're all still done to computers, they just explain it perfectly. I think it's, you don't really get an opportunity like that once you start working to learn everything at a theoretical level.

So, I think people should be studying whatever academia has produced when they're in school, whatever, theoretical tools and not, not brain users.

Steve Hsu: Yep. I also think, uh, for a kid who wasn't interested in trading from the get go. You know, you can imagine they're studying all this stuff. They know all about four A transforms and algorithms and stuff like this.

And then when they end up in your world, they have to basically figure out this whole stack, like what is happening? Like they might not really know anything about T-C-P-I-P or whatever they, they have to basically build it. And I would imagine that's a significant learning curve for even very smart people.

Max Dama: Yeah, definitely. Yeah. So I think if people are learning brain teases, then it's taking time away from something else. So it's almost like gaming the interview process where it makes the interview have less signal and they show up with less skills like you're saying. So, but yeah, it is a very steep learning curve to join the industry.

So I think most companies essentially have a one year training program. They might call it different things. Like they might call it rotation, they might call it something else. But yeah, they'll just kind of teach you how, how things work.

Steve Hsu:

Maybe let's switch gears and talk about ai. So, at this micro level that you're working, do you see large language models playing a big role?

Like, are they already playing a big role? What, what do you forecast for the future?

Max Dama: Yeah, there's few applications. So for large language models, it could be used for either parsing news or text data, or it could be for programming assistance. So I think as far as programming assistance, I think people use them similarly in trading as any other programming field. So they're good for some things, not good for other things.

I, I think, yeah, I don't think I have anything particularly novel to add on, like kind of the coding tools stack.

Steve Hsu: I, I think for funds that have a longer timescale than where you are there's a lot of people searching, like, like thinking to themselves. Like, come on the trading firm of the future of our style.

You know, it is bound to be like, AI is gonna take over more and more of it from human traders and stuff over time. But, I think they struggle to figure out exactly, you know, how to, how to actually get value out of large language models. I think it's just a question that the industry's focused on.

Doesn't, nobody knows the answer.

Max Dama: Yeah. If we're talking about textual data sets that are being used by the trading system, then, then you're kind of in the space of alternate data. So alternate data is like a big research area for anybody who's predicting a couple of hours or longer. So basically when, when you're trying to predict, just say one minute in the future, then the.

The only thing that updates every minute is the tech data. So just trades and ads and cancels. There's no real text data that's updating that frequently. So like, you know, Twitter obviously is updating more frequently, but the actual relevant tweets might only happen once per day.

none: Mm-hmm.

Max Dama: Per company.

So when you're trying to build a data set and you're trying to say like, okay, we need X number of samples to actually fit a model, then you don't have enough samples to fit a textual model in high frequency. and then as far as alternate data for longer term traders, there are, yeah. Analyst reports and also kind of like news data.

Those are probably the two main ones that people are trying to use LMS for. So those are a subset of all the alternate data sources. So all the alternate data sources would include, you know, everything financial, like companies borrowing or shorting. I can't really list all the alternate data sources 'cause a lot of 'em are proprietary and I'm kind of too scared to start listing them.

No, don't, don't do it. Like, do it. What are we, what are we interested in?

none: Yeah.

Max Dama: But yeah, as far as news goes, so news tends to be often echoing whatever's happening in the market or coming from other data sources. And maybe those other data sources are numeric. So you'd rather just go straight to the numeric data and not go through the language part.

And then, yeah, analyst reports. I think people do find value in LMS for those things. So I think for statistical arbitrage or quant systems, you, you're looking for a pattern that's repeating a lot.

Whereas when, when humans read an analyst report, they're thinking, okay, what part of this is not repeating? What part of this is novel or unexpected? And yeah, so that's not really what, quant models are trying to do as much. But yeah, that could be a good application for them. I don't think, I don't think I've heard of anybody doing that. Like really trying to, you know, build a world model and like try to think, okay, what is this company's prospects the next five years vis-a-vis their competitors? And, uh, how the industry's changing and what business decisions they're making. I think, yeah, most people are probably using it more like a sentiment classifier still.

Steve Hsu: Yeah. So that, that's my impression is that that's kind of an obvious thing. Like it can read the news faster than you can, but I think people have this idea that if it is really AGI or ASI in the future, like eventually it's gotta make a better trader on longer timescales like hours or days than, than a human.

But people can't quite figure out how to get there.

Max Dama: Yeah, I mean, I know, I know a bunch of people that work kind of as analysts at banks or in hedge funds. I mean, they used to be banks, now they're hedge funds. So yeah, they'll kind of know one particular sector and they'll say, okay, this sector, the key performance indicator for this sector is. The number of sales or number of big customers or their, their currents of the customers.

And then they'll call all the CEOs and they'll ask like, what do you think about your competitors? And they'll visit the factory and they'll check, okay, you know, this, this company actually listed this address or is it a phony address. I mean, some of those things I think the AI could do, some it couldn't. Like, I don't think that I could currently call the CEO and like, yeah. So maybe

Steve Hsu: Eventually.

Max Dama: Yeah, eventually.

Steve Hsu: Yeah. So, so Max, what is a, what is a day like for you? Like, you get up in the morning, what time do you have to get up?

Max Dama: I guess every company's different, so I could maybe talk about headlines, although it could be different at different companies. So for us, we have a pretty flexible schedule. We just are doing research on historical data, and we're all incentivized to work as hard as we can. So. Some people like to wake up late, some people like to wake up early.

I tend to wake up early. But yeah, then you just start to, you know, come in, check your jobs from overnight, see what failed. Maybe restart something. Maybe think about what experiments to run next, and then send some emails. You know, the market's trading 24 hours, so you can always kind of see what's happening there, but we don't really intervene or, or touch it at all.

So we have the operations team that's kind of in charge of monitoring the trading. And so researchers don't have to do that. That's definitely different at some other companies. Throughout the day, I mean, it's just doing research. So when you're in the flow state, then maybe you're coding something large or you're analyzing data, and then when you get near, at the end of the day, you start to think, what am I gonna run overnight so I can make use of all the cycles?

But yeah, there's no, there's no real routine. I mean, it's, it's kind of like you're, you're building a large technology. So the way I like to think about the career is like everyone is building a technology vertical, and if you can make the technology better than the competitor, then you'll have an edge in the market.

So, you know, maybe at Google, just to make it concrete, like at Google, this could be building Google maps, or it could be building some backend that makes them able to do data or do pull requests more efficiently or something like that. So in trading, you know, that could be like, how do you store data or some type of alpha or some type of risk management system.

So if you have a unique idea on it, then it might take you one year to kind of become an expert on how the, the company currently does it, and kind of just know how to operate it, know how to run experiments on it, know how to make iterative improvements, and then it might take you. Another year to kind of become a thought leader on how it might work.

And then you, you start to build, you know, the new version that kind of pushes forward the frontier. That could take a few years. Maybe you start to manage people because you need extra help and you start to hit some things that you don't have the right skillset for. So maybe you need to work with some developers or some other quants that kind of fill in your skillset.

And then ideally, eventually you kind of build this particular piece of technology to be the best in the world. And then you're trying to think what to do next and think, how do I get new ideas? So that might be like, yeah, the multi-year career path, close to the day to day. But everything in the day-to-day is always similar.

Like its similar cycle, like show up, check the experiments code, look at data launch experiments.

Steve Hsu: So when, when you described your classmates at Berkeley and how most of them, or many of them didn't really last that long in the industry. Does that mean when you get new people who are, say, fresh out of college or grad school and you're kind of watching them, like, do you like to conclude after a while, like this guy has to be managed out, he's really not appropriate?

Or are they leaving of their own volition, but they could have stayed? Like how, how does that work?

Max Dama: Yeah, I'm trying to think because it changes over time. And also like my cohort that I graduated with as opposed to headlands is completely different, right? So like the cohort that I graduated with, they ended up at all different companies and you know, there, there's a lot of different reasons. So a company might have hit hard times, maybe somebody ended up in a seat where they didn't have much upside or they weren't excited about the problems.

So they could have left for any reason and then maybe they just decided like, oh, tech looks interesting, or they want to live in a different location besides Chicago or New York. All the normal life things. I just don't think there was like a very strong magnetic pull for them to stay in the industry for a long period of time.

And there's so many other things that are interesting to do.

Steve Hsu: Got it. So, it's not like you're deliberately triggering the people that enter your firm over, you know, over some timescale. Like you, you'd be happy to like it if they, if, if they like it, if they like it and they want to stay, there's a decent chance they're good enough to stay.

Max Dama: Yeah if we're talking about headlands, then yeah. We try to make it so everyone can be successful and find a place where they can add value and they can grow. I think that's true for most firms. I think there are a few firms that have maybe like a one year training program and at the end they cut like 30% and, I'm not totally sure why they do that. I think one, maybe more positive explanation could be that when you hire undergrads, then you have to wait nine months for 'em to graduate. So you hire them in September, and they graduate in May. Then you have to train them and sorry, I guess I'm missing a step. So when you hire them, you don't know how many are actually gonna sign because you kind of have to blast all the offers before you know how many are gonna sign.

There's no like, early action or something like that. So they may have overhired, and then also the market's cyclical. So after a year when those people are finally up to speed, maybe they want to kind of rightsize the firm. So yeah, so we try hard not to do that because like, it could be bad for the culture or bad for morale, but I think some other companies might have a structure where they kind of have less problems with people leaving.

I think for us, because we're so collaborative and, you know, we all work together, so we don't, we don't really want to have people leave.

Steve Hsu: So you mentioned like eventually all the successful firms in your space, they become basically trading their own money. In that case, I guess AUM is really not the metric you guys care about, right?

It's basically just what's your return rate?

Max Dama: Yeah, I think it's, there's two things people care about. So one is the P and L, so that's kind of like the trading profit after the cost of trading, but before the bonuses and other expenses and the market share. So like, are you number one, number two, and number three on a given exchange.

And most exchanges publish the rank of each participant each month. So it's anonymous, so you know your own rank, you don't know the other company's ranks, but you can kind of back out who's maybe number one or number two.

Steve Hsu: Got it. Now my picture of a, like a more traditional, sort of longer duration hedge fund is, you know, most of the comp is going to a limited number of PMs who, you know, are killing it.

They have a P and L attached to their name and it's huge. And so they get huge bonuses or maybe there's some famous guy who founded the firm who can raise a lot of money from the LPs. For you, like what determines, like who is the quant in HFT, who, like, this guy gets credit for this chunk of the P an L and like, we're gonna give him this much money.

Like how does that work?

Max Dama: Well, like I was talking about earlier, I mean, we're building this technology so you could just think about it like any other technology firm. So I think, yeah, like probably whoever at Google created Google Maps got paid really well, and then maybe some less successful product, like they don't have as strong of a case, something like that.

But yeah, I mean, it's a pretty small firm and all, all the firms in insurance are pretty small. So people know kind of who is the most prolific and breaking through the hardest barriers.

Steve Hsu: You have a way internally just on the team to figure out like who's contributing the most and who, who deserves the biggest comp?

Max Dama: Yeah. I think one thing, like when you're not in the industry, things seem very opaque. But when you're in the industry, like everything is very, very, very transparent. So like, for example, we can backtest every change, and so we kind of know how important every change is and some changes. There's also intermediate metrics.

So if you're building a new alpha, then you can measure, uh, you know, the correlation to the future returns. And so it's crystal clear, kind of like what is significant and what's not significant. Occasionally there's a change that's kind of like very orthogonal and kind of like, but I mean those things maybe after a year or two, then you have the P an L, so then it's crystal clear.

Again, maybe at the time it was added, it wasn't totally clear. Like, you know, say if you're changing the market impact model, then you don't know for sure whether it's better or worse. But you'll see after a year. I think similarly, like when you're not in the industry and you hear alphas, it's like, oh, it's something very secret, something very complicated, very esoteric, and like of course they are very secret. But in practice there's so much alpha, I mean, because, you know, like, like you can find some alphas published online. Like for example, you know, when gold goes up, silver goes up, right? So that's an alpha, and you can build a model for that. And you can basically make that model as arbitrarily complicated as you want.

And it could still make sense because, you know, sometimes they both move up because they're precious metals. But sometimes, you know, maybe silver's more of an industrial metal, or maybe the mine in one of the regions produces both gold and silver. And in a different region it only produces gold and doesn't produce silver because like, however, you know, the geography works.

Or maybe you're looking at the, you know, micro scale. So it's like there could be effects around gold being traded in Chicago, London, Japan. So you could look for effects there. You could look at how the order books are different, so they have different tick sizes and things like that. So you can basically make those models arbitrarily complex.

And that would just be one kind of alpha theme. There's many other alpha themes, like, you know, the main categories would be supply and demand. So when people are buying, they'll typically keep buying. If you're looking on a, on a microsecond scale, you know, it's like if you, if you wanted to predict the sales on Amazon, like you probably have a heavy weight on whether there's a sale recently.

And you know, so if there's a sale recently, maybe it's Black Friday, right? So then maybe people are racing to buy things. And you see the same thing in the market. Like if, if somebody buys, probably soon after somebody will buy. And I mean, that Alpha's been published by probably like Irene Aldridge or other people in academia.

But you can make it arbitrarily complicated. So, you know, the Alpha model would be like adding different features or trying to use better statistical tools to estimate the signal from the noise better and make it generalized further into the future or over other market conditions.

Steve Hsu: But I think what you're describing, while it's very logical for like someone with a quantity kind of, bent, like it's very different from, you know, a more traditional say equity hedge fund.

You know, some guy has a big year and there's always gonna be this question like, was Joe lucky or does Joe really, really have alpha? Right? Like, I, I think even after five years at the firm, they could just be like, this guy just caught one trend and made a shit ton of money and he doesn't really have alpha.

I think there's some kind of irreducible uncertainty about the whole thing.

Max Dama: Yeah, I don't want to downplay like how well they understand their business because like the closer you get to, to a certain business, then you always find out that they kind of know what's going on and it's not that trivial to just like enter it and tell them how to do it better.

And so, yeah, I mean, I think because I'm in HHT and then NIC, oh actually HHT is very, very clear. So I can see standard, I mean, standard, they have a similar kind of alpha setup, and so they might have tens of thousands of alphas because they don't have the latency ceiling, and they can measure on a day-to-day basis the P and L of every alpha.

none: Mm-hmm.

Max Dama: And so there's definitely more variance because like maybe their alphas kind of the central limit, then only kicks in over the month scale instead of the hour scale. Yeah, they can tell very, very clearly kind of like which Alpha worked, which alpha didn't work.

Steve Hsu: How about this, like the Alphas all concentrated in, I bought Nvidia and I've been holding it for the last three, three years.

Max Dama: Yeah. I mean, we're not doing that. Yeah. We're trading a lot.

Steve Hsu: Yeah. Yeah. No, you're at, you're at this sort of hyper-rational limit of all this stuff compared to, you know, more, more kind of big equity hedge funds and things like that.

Max Dama: But yeah, I don't invest in those types of funds because I can't tell whether the managers have skill or not.

I've invested with, you know, say a guy I knew from Berkeley who's running a hedge fund and I talked to him about his thought process and it's very, very consistent and it matches kind of how I think the markets work. And I think his research process kind of works, you know, even if they're hurt and set out firms below him.

'cause I think he's doing something that kind of builds on top of that. So yeah, I basically think of the whole market as one system for pricing. So it's like a pricing computer and each layer is responsible for different things. So, you know, the HIT layer responsible for inventory risk and doing short term price adjustments and then set up layers responsible for doing maybe second order adjustments and propagating the price changes out to like large, large universes.

So if Warren Buffet is buying Geico that he's not gonna place a microscopic trade on all the other insurance companies. That kind of reflects what fraction of his view on that company is that company versus that industry. So you need some trading firms to kind of propagate out those views to all the related instruments.

But yeah, like my, my buddy from college, I think, you know, he's, like I said, calling the CEO, he's visiting the factory and so he's adding exactly the new information that is needed at that layer. And

none: Yeah

Max Dama: He's expressing his views over the time horizon that I think they will realize. And yeah, of course he's had a good track record too, which, you know, always makes it feel better, but

Steve Hsu: That's great.

Well, that's a great, , kind of overview of the, the different, timescales and size scales of the, the whole industry, the finance industry.

So maybe we can close out with advice that you have for young people like my son who wants to go into this field.What's your advice for a college sophomore or junior who wants to go into this field? What should they be doing?

Max Dama: Yeah, so we already went over the academic classes. I think they should apply for an internship. It will help if you do an internship at a tech firm because tech firms generally have better kinds of coding practices and better systems and, uh, tech firms you can work on real problems.

Whereas a lot of trading firms, if they have internships, they're kind of doing toy problems because they don't wanna share

none: mm-hmm.

Max Dama: the real IP. So yeah, do an internship or two, maybe an internship in tech, an internship in trading. Try to build a system on your own. From what I hear from the college students, I think Calci is pretty good.

They have no fees, and it's kind of a level playing field for big participants versus people in school. Like you're just connecting over the internet so there's no edge in higher connection. Yeah, maybe do calci. That's about it. Yeah, just study hard in school and then apply when you get near graduation.

But I mean, I think there's a lot of other areas in the economy that are good places to work too. Like the AI boom is really interesting. So I think yeah, just do trading if you're really passionate about trading, but otherwise there's a lot of interesting paths.

Steve Hsu: You know, his dad is telling him he should work as an AI researcher, but you know, the kids never do what their dads tell them.

Max Dama: Yeah. Usually the opposite.

Steve Hsu: Great. Well, Max, I really appreciate your time. I mean probably this hour of your time is worth a shit ton of money, so really appreciate it and I'm sure the viewers, who are interested in this subject got a lot out of it.

Max Dama: Yeah, I hope so. Yeah. Thanks for having me.

Steve Hsu: Great.