How the Speed of a Trade Got Down to Nearly the Speed of Light

This is Special Agent Regal, Special Agent Bradley Hall. The time is approximately 11.15 a.m. About to start a consensual telephone call with Dr. Daiwa Zhang. China's Ministry of State Security is one of the most mysterious and powerful spy agencies in the world. But in 2017, the FBI got inside. I've never seen that much evidence in my entire career, and I don't think we'll ever see that much evidence again. I now have several terabytes of an MSS officer, no doubt, no question, of his life. And that's a unicorn. This is a story of the inner workings of the MSS. and how one man's ambition and mistakes opened its vault of secrets. Listen to The Sixth Bureau from Bloomberg Podcasts starting on February 13th on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts. Bloomberg Audio Studios. Podcasts. Radio. News. Hello, and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. Tracy, one of the things that I think we like to do on this podcast is sort of de-abstract the things that we take for granted in the world. Right. There are various processes. We always say this when there's like a blow up or something like that, where it's like, oh, if we're having to pay attention to X or Y, there must be something going on. But we don't always have to wait for blow ups. But we live in this world where you click buttons and things happen and you have some intuition of what happens after the button is clicked. But you don't really have a great intuition of what happens between the button clicking and the thing happening. Absolutely. I actually don't know how the process of like inputting an equities trade actually works, like where it kind of shows up. So that's a big question. I suspect a lot of people, even people who are in markets, probably don't know like the entire sequence of events, partly because it's gotten more complicated over the years with like Reg NMS. Do you remember that? Yeah. Stuff like that. The other thing I've been realizing about trading, obviously the big trend here is high frequency trading. Yeah. Right. And it's just getting faster and faster and faster. When we first started writing about HFT, I guess in the sort of like mid 2000s after the financial crisis, I remember thinking that it was all about the actual algorithm and finding like a really smart pattern in financial markets to exploit. But the more I learn about it and the more I read about it, I kind of realized it's not really that. It's about exploiting the market structure. Yeah, yeah, totally. And there's so many, you know, we had that, I forget who we were talking to recently. Oh, it was the guy from Hudson River Trading. And, you know, there were the famous like wire wars where it's like, no, I want to be one inch closer to the main server, etc. It's like, God, this is like a good use of brain power. So like, we're going to solve the market. One more nanosecond faster. One nanosecond faster, et cetera. But, you know, the other thing, you know, sort of related to this, one of my longstanding questions is, you know, a jobs report will drop at 830 on a Friday and the market immediately moves. And I'm like, how did that happen? Because it didn't happen because someone was staring. They had their like fingers above the buy or the sell button. But also had something had to be programmed such that that data could instantly be ingested. And then some sort of like directional trade was made based on that. But I don't know how that happens. I don't know how that works. There's also, of course, the overall question of what all this electronic trading actually means for the market itself. Yes. And people talk about things, you know, with the multi-strap funds, getting these feedback loops and maybe increasing volatility in the market and things like that. So we should discuss. And now I just had one more thing. You say, what does it mean for the market itself? What does it mean for society itself that someone's effort is being placed on like getting a meter closer to the server room or whatever? And why is this a good use of time? And is this improved capital allocation? I'm really excited to say we do, in fact, have, I think, truly the perfect guest to talk about this. Someone who has a sort of an extraordinary body of life's work in a range of areas that is very distinct from almost any other academic or researcher that I can think of. We're going to be speaking with Donald McKenzie. He's a professor of sociology at the University of Edinburgh in Scotland. And I first came across his work. He wrote a fantastic book called An Engine, Not a Camera, which is sort of about finance and the original birth of quantitative finance and the use of advanced models and how these models didn't just reflect what was going on in the real world, but how the adoption of these models then created this feedback loop, the engine effect, such that it actually started to drive markets themselves. More recently, he wrote a book called Trading at the Speed of Light, all about high-frequency trading. He's also written recently a book about digital advertising. And so truly a polymath in the world of thinking about this relationship between industry and the sort of technological substrates that drive them. Professor McKenzie, thank you so much for coming on OddLabs. Well, thank you very much for inviting me to do that. Absolutely. Why don't you start off by telling us the gestalt of your life's work? What is your core underlying interest such that it's produced books in these various realms? Yeah. I mean, fundamentally, I'm a sociologist of technology. So I'm interested in the technical systems that affect or could affect all of us. So over time, first major project in that area was on nuclear missile guidance technology. Then I moved on to safety critical computing technology, then the work on financial models that you've just mentioned, then high frequency trading, then digital advertising, you know, because as well as driving us all insane by ads that we don't want to see appearing on our screens. That's also, of course, the big funding source for much of the everyday digital world. And then most recently of all, I've started working on AI and large language models. So you can see the picture. They're all highly technical areas. One way or the other, they all affect all of us. The other reason we wanted to talk to you is because you come at everything from this sociological perspective. And I absolutely love it when like anthropologists and sociologists go to Wall Street and write about it. Why did you take that approach, especially with your high frequency trading book? Yeah, well, I don't do kind of like quantitative social science. You know, I can leave that, for example, as far as markets are concerned, I leave that to economists. What I do, I like talking to people. I like going, looking at stuff to the extent that you can look at it. I like tracing how things have developed through time. My work's often got us kind of, you know, something of a historical dimension to it. But the most fun bit isn't writing the books. The most fun bit is talking to people. And that's the bit I've always enjoyed most. One interesting thing in this book, Joe, I don't know if you noticed, but Donald writes down like all his numbers of sources and who they are. So like, you know, people from the exchange, people from high frequency trading firms, what their seniority is, which is something I hadn't really seen before. It is a really cool thing. By the way, as Donald says, the fun part is talking to people, not so much the writing of the book. As two people who talk to people every day and have never written a book. I feel already now, granted, we never actually went through the process of writing the book because the talking part is so much more fun. I don't want to ever take a pause from the talking. So I already feel like to some extent, Donald is a kindred spirit. It's like, it's fun to talk to people, isn't it? Talk to us about how you find them. You know, it's like, okay, HFT is interesting to you. And like, you just want to have some conversations. What is the sociologist's toolkit here for knowing who to talk to? Yeah, well, it's always difficult and it's always very ad hoc. There's always a lot of luck involved in it. And with a financial market topic, I will typically start reading the Financial Times, finding names in the Financial Times, approaching those people, and then maybe they pass me on to other people. But, you know, there's also, as I said, dumb luck involved, like a crucial moment in the work I did in high frequency trading was going to interview someone at the start of it. And framed on his wall was the front cover of an issue of Forbes with the headline of the article, Free Enterprise Comes to Wall Street. And I thought, oh, that sounds kind of interesting. And I checked that out and it was to do with a new electronic stock exchange called Ireland. And it turned out that Ireland, the story of Ireland was completely interwoven with the story of high frequency trading. Before we get into exactly what high frequency trading is and how it fits into placing orders for equities or futures or bonds, I have a cultural question, which is whenever you go into an HFT firm's office, it always looks like a tech company. Yeah. With the chessboard and the matcha on tap. And very modern. Why do you think they've taken that approach? How did that aesthetic become the norm? Yeah, that's a really good indicator of cultural change. Because, of course, previous to that, the sort of dominant image we might have of a financial market would be the trading floor of the New York Stock Exchange, you know, folks in colored jackets. They're typically televised when even nowadays when, you know, so there's been a big drop in the market or something. So the cameras try to catch somebody who's looking kind of glum and worried. speed. So we think of that as what finance is. Or we think about like the Bud Fox and Wall Street of a bunch of guys and slick back hair sort of, you know, on the phone. Boilers. Yeah, yelling at each other, looking at a green screen. Sorry, I didn't mean to interrupt that. But those are, I suspect, the two things people imagine. Yeah, yeah. And, you know, there was a transition involved. By and large, the high-fugancy trading firms hire people who know how to code, often, you know, with higher degrees in mathematical kinds of subjects. And even the people who refer to themselves as traders often have that kind of background. And, you know, I'm sure when the visitor isn't there, there'll be a fair bit of swearing at the screen when something goes wrong and that kind of stuff. But you're right that the normal experience of those trading rooms, they're quiet, they're orderly, and you could indeed mistake them for a Silicon Valley startup. Yeah, and you see people in jeans. I visited one, and I think I saw the CEO, and he was just wearing a college T-shirt or something like that. It's a strong memory I've got because in the previous work, they worked for an engine, not a camera, and some follow-on stuff. I would often go to investment banks. And in investment banks, I kind of had to wear a suit and a nice shirt and a tie and so on. So when I started interviewing in high-frequency trading, I turned up at one firm dressed like that. And the owner of the firm sort of snarled at me, you're overdressed. Wow. You mentioned Ireland. Why don't you tell us that story? I want to get more into the tech, et cetera, but you're like, okay, this turned out to be an exchange. What was distinct? What is Ireland? I've heard of it, but it's again, one of these things that I've heard of it and then I moved on. What was distinct about this and why is it so interwoven into the history of HFT? How is it different from other exchanges that have existed for hundreds of years? Yeah. Yeah. I'm going to oversimplify, of course, because there were predecessors to Ireland, you know, were a little bit like it and so on, but that would take us too long to go into. I mean, And fundamentally, trading on Ireland was organized around an electronic order book, which is a list of all the bids to buy or offers to sell the shares in question. And that electronic order book is managed by something called a matching engine. And as the name implies, that looks for a match. In other words, a bid to buy and an offer to sell at the same price. And when it finds that couple, it consummates the trade and the trade is done. So it's all done electronically. There's no direct human negotiation involved. You just enter your orders into the order book and the matching engine either executes them or fails to find a match. There were exchanges prior to Ireland that worked in that kind of way. But what was distinctive about Ireland is that its matching engine was blisteringly fast by the standards of the day, which was essentially the late 1990s. So the closest analogue was a system called Instanet. And it might take a couple of seconds, the matching engine, to find the match and execute the trade. And of course, for a human being sitting there, even if they're impatient, two seconds is not a very long time. Ireland improved on that a thousand fold. So it could execute trades in two milliseconds, two thousandths of a second So that was the opening for high frequency trading that with exchange I mean, strictly Ireland was not exchange, it was what was called an electronic communications network or ECN, but I'll call it an exchange for simplicity. If you've got an exchange like that, and you've got an automated trading system, it's a marriage made in heaven. The two things, the exchange and the trading firm fit each other very, very well. And amongst the consequences of that is that liquidity in Ireland, it traded NASDAQ stocks. And this is the time of the dot-com bubble, of course, where there's a lot of trading of NASDAQ tech stocks. Ireland brought a lot of liquidity to that market. So that's, you get a kind of feedback loop where you get automated trading, bringing liquidity to exchanges that have the kind of technical features that make high frequency trading attractive and feasible. So the established exchanges started to have to change how they did things because otherwise they were going to lose out to the new exchanges. and that's basically the feedback loop that's created today's electronic markets. We're up early every weekday keeping an eye on what's happening across Europe and around the world. We do it early so the news is fresh, not recycled, and so you know what actually matters as the day gets going. From Brussels, I'm following the politics, policy and the people shaping the European Union right now. And from London, I'm looking at what all that means for markets, money and the wider economy. We've got reporters across Europe and around the globe feeding in as stories break. So whether it's geopolitics, energy, tech or markets, you're hearing it while it happens. It's smart, calm and to the point. And it fits into your morning. You can find new episodes of the Bloomberg Daybreak Europe podcast by 7am in Dublin or 8am in Brussels, Berlin and Paris. On Apple, Spotify, YouTube or wherever you get your podcasts. Jo, whenever I hear terms like millisecond and nanosecond, I just, it's so hard to wrap my head around what that actually, It's faster than that, for sure. I can actually help there. I'm going to hold my fingers. I'm holding them 30 centimeters apart, or since you're in the U.S., I'll say one foot apart. At the speed of light in a vacuum, it takes a nanosecond to get from one finger to the other finger. And that's an indication of how fast automated trading, specifically high frequency trading, has become. That when I started working on the topic in roughly 2011, people were still talking about milliseconds or thousands of a second. Two or three years later, it had become microseconds or millionths of a second. And by the time I was finishing the research, nanoseconds were starting to account. So light traveling that 30 centimeters, traveling that foot, that mattered to high frequency trading by roughly 2018, 2019, 2020, round about then. That's very helpful. I do have questions about the physical realities of how fast we can actually go with all this stuff. But before we go any further, can you talk about the process of let's just focus on the equity market for now. Someone places an order to buy or sell a stock. What actually happens in the ecosystem between traders and market makers and the exchange that makes that happen? And what does it mean to actually make that happen and execute the trade? So what happens is your order via the broker, your user, I mean, in the brokerage system is no longer a human being, via the brokerage system gets placed in the exchange's order book. And then one of two things happens. The first is if the matching engine can find an existing order in the order book that matches the price of your order, it executes the trade. And the trade then happens not quite instantly, but, you know, very, very, very fast. And you're done. That's it. It's over. If, on the other hand, there is no match as the order book stands, your order rests in the order book. And it stays there until either you cancel it or a matching order comes along and then it's executed at that point. So that's the basic process. You know, so it occurs to me like gains of speed in trading have been happening forever, long before we were talking about, you know, anything electronic. I'm sure other technologies exist. Technological evolution is a long time thing. It is a pretty banal statement. I suppose. But what I'm curious about is the sense of which a change in degree becomes a change in kind, essentially. So that like when you go from one second to a thousandth of a second to a nanosecond, how does that change, say, like the types of strategies that can then be employed or the types of skills that might be required to be a successful trader in the nanosecond era versus the one second era? Like talk to us about like that relationship. Yeah, yeah. Well, there's a wonderful book by the historian of technology, Jimena Canales, which is called A Tenth of a Second History. And the significance of the tenth of a second is that's the generally accepted lower threshold of the human perceptibility of time. You know, basically, we just can't mentally process time intervals that are less. So, Tracy, don't feel bad about not being able to build an intuition for a nanosecond. Less than a tenth of a second. So what essentially happened is that we've moved from that kind of, you know, the tenth of a second or longer, from that kind of epoch into an epoch where human beings, I mean, they can still be in overall control of the system, But they can't actually execute the actual trading decisions fast enough not to be outrun by an algorithm. So we've moved from a kind of human-centered form of trading to a machine-centered form of trading. And the actual threshold of the change is probably around that tenth of a second amount. amount. I have another cultural and I guess market structure question. But one thing that I always thought was interesting about high frequency trading was that the banks didn't really get into it, which, you know, there's one big reason why, which is the ban on prop trading after 2008. But even before then, they just never seem to be able to compete with independent firms. Why did that happen? Yeah, it's I've asked people that. And there's, I think, complex thoughts of reasons. And let's let it be said, some banks have been more successful than others. Banks are not always bad at this, though most banks are bad at it. One way of thinking about it is that typically a bank will have an IT department that separates from the other functions of the bank, like trading, like market making and so on. So, you know, if you're a trader, you got to persuade the IT people to give you a fast enough system, which involves them, you know, maybe writing some new software, buying some new kit. So you need to get higher level management sign off on it. And it all takes time. Whereas the high frequency trading firms are typically pretty small. You know, 50 people is a decent size firm. 150 people is, you know, is a reasonably big high frequency trading firm. Very often those firms are owned and run by the people who founded them. So there is a boss or bosses. But other than that, it's a relatively flat organizational structure. If, for example, at least this was the case in the early days of high frequency trading, It's not quite as simple as this now, but in the early days of high-frequency trading, you know, if some IT firm came out with a new, better, faster server, and you were a trader in a firm like that, and you're like, okay, well, let's get this new server. you could just use your own personal credit card to buy the server, get it delivered to your office, and then get your engineers to take it out to whatever data center that they were trading in and get it installed straight away. And, you know, that could maybe that would be a week or 10 days or something. Whereas in the bank, you'd be doing pretty well if you could achieve that within six months. Yeah, that's amazing. purchasing credit cards. Tracy, we don't know anything about long waits for computers to arrive. No, we surely don't. That's sarcasm, by the way. Actually, this reminds me of something that I wanted to ask, which is we know there's competition between firms, high-frequency traders, for the fastest connections to exchanges and things like that. There's also competition, I imagine, within the firm itself, because setting aside the credit card anecdote, these can be expensive and there's also limited supply. You know, only so many servers can be co-located where they want to be. In your research, how did you actually find executives at HFT places? How did they actually allocate the fastest connections to which team? What I found was the high-frequency trading firms fell into two different camps, so to speak. In some of them, there were separate trading teams that didn't really communicate with each other, and indeed by design didn't communicate with each other. In some cases, the office was actually laid out in such a way that somebody in one team team was not very likely accidentally to overhear something said by someone in another term. And in those firms, yes, I mean, they are essentially in competition. And I think that in that kind of firm, then the results of each trading desk, the P&L, the profit and loss, the little trading teams that are doing best would get the available bandwidth on the microwave links that are crucial to high we can see trading and so on, and maybe they would get the fastest machines first and so on and so forth. The other kind of trading firm was and is operated as a unified entity. In some cases, even without individual profit and loss in individual P&L accounts for traders. And there's a lot of shared infrastructure in that kind of firm. And indeed, there's also shared infrastructure in the segregated kind of trading firm? Because if you're the boss of such a firm, there's obviously simple economies in not having completely separate IT systems for each trading team. But there's that kind of divide. Does the firm operate as a unified entity or does the firm operate as a sort of aggregate of competing trading teams? So actually, let's just, you know, on the subject of who is the closest or who gets to have their server located where. Tell us a little bit more about the timeline. So Island emerges in the late 90s. When did it start to dawn on people in the trading industry that given this new physical reality, given the speed, we need to start thinking about who is going to have co-location? We need to start thinking about sort of like microwave radio line of sight. Where did that speed war? What was the genesis of it? Yeah, yeah. I mean, a kind of crucial date was 2005, where Ireland, which had already been bought by Instanet, was acquired by NASDAQ. And the New York Stock Exchange acquired another of the pioneering electronic trading exchanges. This one was called Archipelago. Is that just a coincidence, that island and archipelago? That's kind of like... Yeah, no, I'm sure it's not a coincidence, yeah. And they technologically reorganized themselves in New York Stock Exchange and NASDAQ, technologically reorganized themselves around this new insurgent technological approach to trading, so to speak. And 2005, because of the acquisitions in that year, is a kind of noteworthy year. But even before 2005, people in trading firms started to become aware that you couldn't just do automated trading with the machine sitting in your office. For example, there's a Kansas City trading firm called Tradebot, whose owner Dave Cummings has written a really rather nice autobiography. And one of the things that Cummings came to realize is that trading an island while having your machines in Kansas City was placing him at a disadvantage So firms like that started to move their machines either directly into the island computer room or they couldn do that into the offices of another firm in the building to shorten the distance And that kind of thing was already in place by 2005. Two things then happened. For a period, there was a kind of Wild West, so to speak, where there are lots of stories of high-frequency traders like drilling holes in walls so as to shorten the distance between their servers and the exchanges matching engines. That has by and large generally come to an end. And what happens now in the data centers of all the major exchanges is there is a rule about equal cable length. So if you happen to have, if a trading firm happens to have its servers physically close to the exchange matching engines, the fiber optic cable that connects them is coiled so that there's exactly the same cable length for each of the trading firms within that. data center. The other thing that started to happen is that getting the signal from one exchange's data center to another exchange's data center started to become a technological speed race. Back to 2005, by and large, it would just be sent by fiber optic cable, but the exact route was not under the control of the exchanges or of the trading farms. So there's a lot of sort of randomness. The crucial link here is actually between the Chicago Mercantile Exchanges data center. Then in downtown Chicago, it's now in the suburbs of Chicago. the link between that and the data centers that trade shares in northern New Jersey. And there was a kind of triple evolution there. The first evolution was that the particular trading firm, and since it's no longer directly in business, I can actually name it Getco, managed to, as it were, stitch together existing fiber optic cables to get the fastest route on the Chicago-New Jersey link. And that actually in the business was known as the gold line. Gold because of the money that you could make by having the fastest route. Then in 2010, if memory serves me right, a new firm, Spread Networks, actually dug an entire new cable from Chicago to northern New Jersey. you know drilling you know sort of underneath car parks and you know just really trying to be as close to what a geographer would call the geodesic in other words the fastest route on the surface of the earth from point a to point b then third phase that was trumped by the arrival of microwave because lighten a fiber optic cable i mean the core of fiber optic cable is essentially a specialized form of glass, and that glass slows the light down to only two-thirds of its speed in a vacuum. Whereas if you can shoot your electromagnetic signal through the atmosphere, it's not exactly at the speed of light in the vacuum, but it's very, very close to the speed of light in a vacuum. So that was the third phase when people moved from exclusive use of fiber optic cable to supplementing the fiber optic cable by microwave links between Chicago and northern New Jersey. That was incredible. The news doesn't stop on the weekends. Context changes constantly. And now Bloomberg is the place to stay on top of it all. Hi, I'm David Gurra. Join us every Saturday and Sunday for the new Bloomberg This Weekend. I'm Christina Ruffini. will bring you the latest headlines, in-depth analysis, and big interviews. All the stories that hit home on your days off. And I'm Lisa Mateo. Watch and listen to Bloomberg this weekend for thoughtful, enlightening conversations about business, lifestyle, people, and culture. On Saturday mornings, we put the past week's events into context, examining what happened in the markets and the world. Then on Sundays, we speak with journalists, columnists, and key political figures to prepare you for the week ahead. Join us as soon as you wake up and bring us with you wherever your weekend plans take you. Watch us on Bloomberg Television, listen on Bloomberg Radio, stream the show live on the Bloomberg Business app, or listen to the podcast. That's Bloomberg this weekend, Saturdays and Sundays starting at 7 a.m. Eastern. Make us part of your weekend routine on Bloomberg Television, radio, and wherever you get your podcasts. I have another cultural slash market structure question, which is, Along with this intense competition to be faster than anyone else, there was a narrative around that time that this was a bad thing, right? And one of the interesting cultural things is you saw the high-frequency trading firms kind of divide themselves into good guys and bad guys. So there were some that were saying, well, we make liquidity, right? We're good for the market. And then there were others. I mean, they wouldn't describe themselves this way, but they were accused of being liquidity takers in the market. How should we think about that particular tension? Yeah, no, that's a very fundamental thing because you're quite right. There's a degree of differentiation between trading firms and indeed in the more segregated firms also within those firms, you have different desks fall on others on different sides of that. The way to think about it is to remember that in all these exchanges, trading is organized around an electronic order book. As I said, a list of the bids to buy and offers to sell the instrument in question. What a market making firm does is it places lots of orders into the order book at prices that can't immediately be executed. But those orders then populate the order book. So if somebody else comes along, an individual investor or maybe an asset management firm, somebody else comes along and needs to trade. They find an order book that's populated with lots of existing bids and offers that they can execute against. So the market making firm is providing liquidity. And most people reckon that that's a good thing. The other kind of firm, the one where, as you say, there tends to be more controversy, they don't do that. They don't constantly populate the order book. Their systems constantly monitor the order book. And then when they detect what they think is a profitable opportunity, they execute against the orders that are already in the order book. And that is called liquidity taking because, of course, the execution removes the order from the order book. And in practice, a lot of this is actually going on between high frequency trading firms, because most of the orders in the order book are placed by market making high frequency trading firms. And a lot of the executions against those are by high frequency trading firms that specialize in taking liquidity. And this is the core of what gives the business its characteristic as an arms race in speed. So imagine for a moment that your algorithms are trading equities in one of the data centers in northern New Jersey and the relevant stock index future traded in Chicago changes in price or even there's a big shift in the order book for that stock index future. And let's say the price of the stock index future falls. That's very likely to lead within a tiny fraction of a second to falls in the price of the underlying shares being traded in New Jersey. And in that tiny little fraction of a second, in that intervening period, a lot of the market making firms orders in those order books become stale as people in the markets. So the making firms rush to try to cancel their stale orders and the taking firms race to execute against those stale orders. And that's a race that nowadays can literally be played out in nanoseconds. That's interesting. I hadn't appreciated that dynamic at all, to be honest. You mentioned that these high frequency trading firms, they're all sort of like they're run and operated basically by the employees, the owner or something like that. What is it that sort of distinguishes a high frequency trading firm from, say, a hedge fund that would have limited partners and make distributions, et cetera? You know, you never hear about like, oh, I have an investment with Jane Street or something like that. I don't think that's the right thing. What is it about the nature of the business such that essentially they trade their own capital? Yeah, I guess that's just in a sense. It's one of those things that's historical evolution. I mean, in many cases, those high-frequency trading firms were initially set up by successful floor traders, particularly floor traders in the Chicago markets, the futures markets. And if you were successful in that, you could, you might not become a billionaire, but, you know, you could make a decent amount of money, tens of millions of dollars. And that was enough to enable you to start an initially small automated trading firm. And you often didn't need any external capital to do it. You just had to buy the necessary technical kit and hire technically savvy people and so on. But you could start really quite small. You could start a 10-person firm or something like that. Now, the business has got a lot more expensive since then because in good part of the speed race that we've just been talking about. But by and large, those firms made profits. The owners reinvested the profits in the firms so that, you know, the capabilities of the firms grew to meet the growing demands on them. And then another thing that should perhaps be said is that those founders would have the majority of their net worth invested in the firm. And so their risk control was often pretty good. You know, risk management for those firms was not a sort of, you know, a separate bureaucratic department that the traders had to try to outwit, so to speak. the founder would quickly detect if you were trying to do something like that. Now, some automated trading firms are blown up nevertheless, but it's actually quite interesting how few of them have blown up. Yeah. Because of, you know, for example, because of classical software bugs. And that's, I think, because the relatively small structure, the hands-on involvement of the founders, et cetera, et cetera, has created a kind of technical culture that actually works pretty reasonably well, better than I would have expected it to work at the start of this research. So one thing I wanted to ask is this idea of physical limitations to how fast we can actually go. I'm pretty sure people always say that you can't go faster than the speed of light. There's probably some caveats there about like quantum entanglement and stuff like that. But surely we must be getting close to how fast things can actually go. What's your sense of how long the speed race can continue? Yeah, well, we can never get to zero, of course. I think Einstein is basically correct. You can't get faster than the speed of light in a vacuum. and similarly a computer system there's always going to be a non-zero processing time of the system but you can get ever ever closer to that so in mathematics speak zero is an asymptotic limit that's to say you can always get closer to it but you're never actually going to get right there. And I think that's the nature of the business. You know, we're still in the nanosecond regime. If I remember correctly, the next lowest time interval is the picosecond. You know, I could imagine this continuing in a domain of picoseconds. So that's the way I would see it, that there's a hard limit, but we're never actually going to get to the hard limit. We're still going to race to get as close as possible. But your understanding as of the time of your work is that the race is not over, that for these firms, and I'm sure they have many different projects on their plate, including various things with AI, which we haven't gotten to, and I guess we won't. But the speed race is not and will never be done. Yeah I think that correct Now of course there is an economic process at work here which is to say that the investments that you make in speed have to be recoupable from the trading profits that you make from your trading. And I think, Tracy, I think said at the very beginning, what essentially is going on here is that structural features of financial markets are being exploited, like the relationship between the stock index future and the underlying equities. The amount of money to be made by exploiting those kind of structural features is not trivial. It's been nicely measured by the Chicago economist Eric Soufert and colleagues. It's not trivial, but it's perhaps single digit billions of dollars. So suddenly deciding you're going to invest $50 billion in the technology of speed would be a dumb thing to do because you wouldn't be able to recoup it. So there is that economic factor that is, I'm pretty certain, slowing. The speed race is still there, but it's, you know, things are getting faster, but the rate at which they're getting faster is certainly not accelerating. And I think that economic factor probably explains it. That makes sense. So we should talk about the impact of HFT on the overall market a little bit more. And one of the things that caught my eye in your book was you cite a previous study. I can't remember by who, but basically saying that the efficiency of financial markets has not improved between the 1880s and 2012, which is very counterintuitive. It seems like impossible to imagine, but what does that mean? Yeah. So that is work by Thomas Philippon, or his French are pronounced in the American words, Thomas Philippon. What he means by efficiency there is really rather different from what I've been talking about. What he means by efficiency is what he calls the unit cost of financial intermediation, which essentially is basically putting it crudely, how much it costs to do the kind of thing that investors want to do, that asset managers want to do, and so on. And, you know, it is a very striking finding that from the 1880s to, I think, his most recent data goes up to 2015, that there was no really clear cut tendency for that cost to decline, despite all the advances in information and communication technologies over those many decades. And the explanation, to put it simplistically and crudely, is the capture of those efficiency gains in the form of high pay in the financial sector, typically through the form of fees. So the fees that you pay for an index fund, say, for example, those have really gone down, but people have also been moving their money into private equity and the like hedge funds, which have much higher fees and so on. So those kind of effects seem to have sort of cancelled themselves out. In the 1940s, a professional in finance was basically paid roughly the same as somebody with equivalent educational qualifications in a different line of business. And then from the 1970s onwards, the gap has got bigger and bigger and bigger now. These days, of course, you can make a lot of money by being a technologist in AI, for example. But by and large, those sort of exceptions aside, finance is an extraordinarily well-paying professional, at least for those in the central roles in it. And that's essentially the explanation of that finding by Philippon. Well, this is a good opportunity to ask about AI because I suppose it's inevitable as a sociologist who examines the tech industry or looks at how tech is impacting things. Your next project must be AI, right? Yes, it is. And what's the particular angle or what have you been discovering so far? Yeah, it's very early days. But the thing I'm most interested in so far is the question of scaling and AI. Because, of course, it's no secret to anybody who reads a newspaper, subscribes to Bloomberg or whatever. I mean, the huge trillions of dollars are being thrown at AI infrastructure. And absolutely, there is a sort of logic there that's repeatedly stated that these systems are all built around neural networks. And the effectiveness of a neural network grows with the size of the network, the size of the training data, the number of parameters in the model and so on. And there are well-known scaling laws. But, and this is the thing that interests me, is the but. There's a very nice little statement from Sam Altman in February of last year that the intelligence of a system is roughly the log, the logarithm, in other words, of the resources devoted to training it, running it, to computation at inference time, and so on. Now, of course, what Altman meant was basically, give the industry more money, and you'll get more intelligence. And that's, of course, indeed, giving more money to the industry is exactly what's going on. But a logarithmic function, there's a bit of maths here a logarithmic function at least of the kind that altman is referring to is a diminishing returns function you can draw its graph and it very clearly demonstrates diminishing returns you can always get better and better but each increment costs you more in terms of the resources deployed. And we're dealing here where the horizontal axis in the graph, so to speak, is denominated in trillions of dollars of financial input or hundreds of megatons of carbon dioxide emitted by the electricity generation needed to power the data center. So the question becomes, on a diminishing returns curve, how far do you go? When do you decide we really got to stop? Can you decide we really got to stop? Do you have kind of quasi-magical beliefs, so to speak, that at some point the diminishing returns, something qualitative will happen? that artificial general intelligence or super intelligence will suddenly appear. So that's the core of what I'm interested in right now. How far do you go along a diminishing returns curve? Joe, I just want to state for the record, if you give me more money, I get more intelligent. There's no diminishing returns. Just for the record. You know, noted, first of all. And it's interesting, you know, hearing this in the context, And it suddenly makes so much sense how this fits into your work and this idea of like the arms race. Right. Because, yes, it's true. Like maybe there's only so much extra profit available for the firm. And maybe that pool of profit is shrinking and maybe it gets more and more costly to sort of exploit the remaining profit that's available. But on the other hand, you can't fall behind. You can't let – and this is – so it's true in HFT and it's clearly true in AI where, okay, like it spends more and more money to improve the model. But you can't fall behind even if the economics look worse with each iteration. Anyway, Professor McKenzie, lovely conversation. I really enjoyed that. We really learned a lot. Really appreciate you coming on to OddLots. And yeah, thank you for joining us. Well, thank you both. Thank you both for inviting me, like I said. And, you know, thanks for a really great conversation. That's fantastic. Thank you so much. We'll have to have you back on when you publish your AI book. Absolutely. Tracy, I love that conversation. Really interesting. I love, like, encountering people who are, like, actually understand the tech. actually can articulate what the tech is doing. Especially, it's always impressive, someone with a sociology background, et cetera, to just sort of be like that comfortable. And I think that's like the through line of his work is like, he gets it. Right. So in the book, there's lots of like field trips to data centers and looking at cables and things like that. But then also, as he stated, just talking to people and getting anecdotes. And there's funny stories about like the battle of the asterisks and things like that. That's at the very end. But people should go and read it. The other thing that stood out to me from that conversation was towards the end when we discussed AI, you made the point that you get this similar dynamic between HFT and AI now where because everything is framed as existential, you just can't stop. Right. You always have to keep going. Totally. Look, I mean, I suppose the high frequency trading is not sort of like existential in the broad sense, but it's existential at the firm level. That's what I was going to say. Yeah, at the firm level. So it's like and I hadn't really thought about, OK, you like have this like pool of theoretical profit, which is the gap between where the futures are trading in Chicago and where the stocks are trading in New York. That's fixed, right? That's not going to get that big. But again, someone gets faster at exploiting that. And I had never really heard quite until your question and his answer, this sort of maker-taker dynamic of, okay, I have these orders and now I'm quickly rushing to cancel them and you're quickly rushing to fulfill them. and if you and I are both in the market, we can't slow, if you get faster, I must get faster because you're going to then snipe me every time or vice versa, et cetera. But, you know, they still make a lot of money, it seems like, unlike the AI firms. They make a ton of money. Yeah, exactly. That was a funny dynamic in HFT world, the like accusations of taker versus maker. And I always think of the, you know, the Spider-Man meme where they're all kind of pointing at each other. It felt very much like that. But it was really great to catch up on HFT again. This was sort of a blast from the past because you used to hear about it more. And now it's become so normalized that people just don't talk about it that much. Well, you know, we heard about it a lot and especially post 2008. Yeah. That was the ultimate finger pointing era, right? The Michael Lewis book. You just have like everyone had some, oh, it's the naked short sellers. It's the credit rating agency. It's the, you know, the law that forces banks to be equitable and who they distribute mortgages to, etc. Like there was a million finger pointing. Oh, maybe it's the HFT firms. Maybe it's the short sellers, whatever. So, I mean, part of the reason we don't hear about it as much is because there hasn't been a crisis, etc. But as you said, the race continues of various flavors. You can never get to zero, but you can always get closer. You know, it's like, you know, I always think about some lines. they go like straight up, you know? It's like, they ever like curve back around, you get like negative space. Like, can we do even better than line go up? Like line go up and backwards? I guess Einstein would say no. If only we could have Einstein on as a guest to talk about trading. To talk about high frequency trading. Shall we leave it there? That would be a perfect guest. Let's leave it there. All right. This has been another episode of the Odd Lots podcast. I'm Tracy Allaway. You can follow me at Tracy Allaway. And I'm Joe Weisenthal. You can follow me at The Stalwart. Check out Donald McKenzie's book, Trading at the Speed of Light. And of course, follow our producers, Carmen Rodriguez at Carmen Armand, Dashiell Bennett at Dashbot, and Kale Brooks at Kale Brooks. And for more OddLots content, go to Bloomberg.com slash OddLots. We have a daily newsletter and all of our episodes. And you can chat about all these topics 24-7 in our Discord, discord.gg slash OddLots. And if you enjoy Odd Lots, if you like it when we look back at the HFT boom and how it continues, I guess, then please leave us a positive review on your favorite podcast platform. And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening. This is Tom Keen inviting you to join us for the Bloomberg Surveillance Podcast. It's about making you smarter every business day. I'm Paul Sweeney. We bring you complete coverage of the U.S. market open. We cover stocks, bonds, commodities, even crypto, all the information you need to excel. And I'm Alexis Christophorus. Bloomberg Surveillance also brings you the analysis behind the headlines. We do that through conversations with the smartest names in economics, finance, investment, and international relations. We do all this live each and every weekday that bring you the best analysis in our daily podcast. Search for Bloomberg Surveillance on Apple, Spotify, YouTube, or anywhere else you listen. On the East Coast, listen at lunch. And on the West Coast, listen as soon as you wake up. That's the Bloomberg Surveillance Podcast with Tom Keen, Paul Sweeney, and me, Alexis Christophorus. Subscribe today wherever you get your podcasts. Bloomberg Surveillance, essential listening each and every business day.