What It Takes To Win At Quant Investing

the face of the biggest economic crisis in decades. But I feel like a lot of people just sort of missed various turning points in the market as well, and are very very frustrated. Absolutely, I mean just super super high, super high levels of frustration. Also, even if you were along this market and sort of like generally bullish, the only way to have really won this year would be

super concentration in tech stocks. And I feel like if you were under exposed to like a handful of tech stocks, which we could count down about two hands, then you're almost guaranteed to be sort of underperforming your benchmark this year, whatever it is Yeah, I think that's absolutely true. And of course we've been talking about for years and years and years that the big Tex stocks, saying whatever you

want to call it, are potentially overvalued. So it's it's doubly ironic that this year you would have underperformed have you not invested in the stocks that people say it might be the most overvalued? Right, And of course that is a big frustration to investors who have been waiting a long time for other sort of factors to do well.

So investors like to talk into factors and this sort of the growth factor has done phenomenally well, but historically the value factor, so called cheaper stocks, those have done well, and everyone keeps waiting for this turn or for other factors to immerge, whether it's value or low beta or something else. Uh, never seems to happen. And if anything, this year did not prove to be a turning point in the market, but really just sort of an accelerant

of it. Yeah, I think that's right. I'm actually looking at a chart from Big of America Merrill Lynch right now, and they point out that values relative performance to growth was the worst this year since the dot com bubble, So um something to remember, But we're not. This isn't this podcast isn't about value versus growth? Is it? No,

it's not. But I think that the frustration that people probably have this year does lead to um, you know, people looking for other approaches to investing, and of course in times like this, people wonder if, like maybe other sort of quantitative or algorithmic strategies more money should be poured into them as an alternative to this ride where you just sort of by the big tech s docks

and I hope that you you know, avoid the turning point. Well, I guess another way of putting it is a lot of the a lot of the quant strategies are sort of momentum based, right, So if you can figure out where the money is flowing to, even if it's tex stocks, Uh, that might be a good way of investing in the current environment. If everything's about liquidity and following the flows, then quant investing or algorithmic trading, whatever you want to

call it, might be a good way forward. Yeah. But you know, backing up, it's like we talk about quant investing and the word quant gets used all the time, and sometimes uh, it's used to describe these super technical funds, and sometimes it gets used to just describe sort of anything that has some statistical analysis of it. That that term feels extremely vague. Yeah, and potentially overused as well. Right, Like everyone wants to seem like they are quantitative in

some way or another. No one wants to say that they're investing purely on emotion and gut feeling and that kind of stuff. So quant gets bandied about quite a bit.

So today we are going to talk with an expert who is ah, knows a lot about quant investing studies that can help us define it and uh also hopefully sort of explain to us what it takes to win in this space, because again, everyone sort of wants to be in the space, even you know, traditional hedge funds over the years have allocated more and more money to quant, to hiring PhD s, to building up their computer systems, But what it really takes to win and can lots

of players succeed is still uh kind of an open question. Yeah, I think that's exactly right. And as we're going to discuss, quant investing is probably one of the most expensive ventures that you can sort of embark on. Yes, Okay, So without further Ado. Let's bring in our guest. He is an expert in the field. He is ciamac Millmy. He is a professor of business. He's a professor at the Columbia Business School, done a lot of research in the area of quant investor. He's also a part time partner

at a fund himself. Thank you very much for joining us. Thanks for having me. I'm delighted to be here. When I say quant investing or when people say quant investing, what does that mean to you? Like, how would you just define that term and so that it's a useful so that it's a useful term. Well, people have different definitions. I personally define it as having two key characteristics. Um. The first characteristic is that the investment process is entirely systematic.

So there's many different times of types of investment strategies that the people implement that employ at some level quantitative methods. But I think the key to the quantitative methods that we're going to speak about today is that at the trade by trade level, there is no discretion. Right. Um, You've you've set up an algorithm, a particular system on a you know, second by second trade by trade basis,

the everything is being automatically done. You know. That isn't to say that there isn't like a portfolio manager involved. But the job of the portfolio manager is not so much deciding on trades and sizing them and so on, but more setting up the computer algorithms in advance and tweaking them and improving them over time. So so that's really the first big component to be entirely systematic nondiscretionary.

The second component of the ones that I focus on is that they're really active investment strategies in the sense that you're buying now because you think the asset will be worth more later it's it's mispriced in some level, or alternatively, you're selling short now because you think the value later will be will be lower. There are other flavors of quantitative strategies that are somewhat more passive, things like uh, you know, exotic beta, investing in factors and

so on. Um, those are not so much a little bit less my area, and I have my own views, and then we can get into later perhaps, But the key things I'm thinking about here, you're using algorithms and data and machine learning and so on. You're taking an active view on what the current prices are relative to what you know the value might be later. So is quant investing proof that markets aren't efficient? I feel like this comes up a lot, but maybe it's worth asking

this question early on. If the whole strategy is to automatically arbitrage price discrepancies in the short term versus the long term, does that mean that markets aren't doing their job? Well? I mean, I think if you want to sort of take the straw man that the markets are, you know, sort of a hundred percent efficient and prices are incorporating

all potential information, I think that's clearly not true. And I think, um, the long term success uh and incredible performance of you know, quant investors like Renaissance is is sort of one piece of that um. But that doesn't

mean that the markets are completely uh inefficient either. LASSA Peterson who's from from n y U and a q R. He has that he has a nice phrase called inefficiently efficient, or I should say efficiently inefficient, meaning that there are inefficiencies, but it's a competitive game and there are lots of smart people with you know, a lot of resources going after these inefficiencies, and when you identify them and trade

on them, Um, they disappear, they're armed away. So you know, these these inefficiencies typically lie around the frontier of the transaction costs, of what it costs to trade. So, um, yes, there are inefficiencies, but they're they're hard to find, and you know, they disappear over time. So one common concept

that quants talk about is is alpha decay. Like you you identify some some signal or some inefficiency and uh, you know, generates a certain amount of P and L and literally year over year you can see that decay away. And you know that's that's because that inefficiency eventually is identified by other people and as more and more people trade on it, you know again it disappears. So it's not that um, you set up an algorithm and it

just you know, sort of prints money. Um uh you know, some sort of gross violation of the the efficient markets hypoth is That's that's not how it works. The people who are successful at this are constantly investing and deploying enormous resources, hiring large numbers of PhD s, and uhum, progressively innovating in order to have new models because because the old stuff will simply stop working. So it sounds like I mean I guess you just said it, but

it sounds like the key to winning. And we'll get more granular in a second. Is that continuous process. It's not about identifying some flaw in the market or some inefficiency or some opportunity to make money. It's about having a team and a process to keep finding those over and over again. That's right again, because all the inefficiencies that I've ever seen are are short lived. So can you maybe um talk to us a little bit more

than about how a quant strategy might be developed. So obviously you have the techno logical aspect of it, the need for computers that are able to trade very very quickly. You have the need for servers, many of them co located close to the exchanges. But then you also have proprietary data sets sometimes and then you have proprietary algorithms. So how does that all come together into one quant strategy? And which one of those is sort of the most

or the biggest investment for a quant firm? Got it? So I think there's definitely a technological investment, may or may not involve things like co location near the exchanges, so at least anecdotally. For example, Renaissance, which is the most successful quantitative firm does not co locate. You know, again, I don't know, but that's that's that's what I've heard.

Co location is quite important when you're trading UH and you require very low latency and and that's typically the high frequency trading domain and which again intersects with with quant in many ways. But if you're looking UM a little bit longer, if your horizons are a little bit longer, it becomes a little bit a little bit less important. Your broader point, I think is correct. Technology is important. I think I'm more important. It's kind of a research process.

There there's a number of kind of high level pieces to a successful quantitative strategy. It's not like uh, UM there's just a black box and UM in gooes data outgoes trades. There's there's a number of pieces in there that UM sort of split the problem into to kind of make it manageable. UM at the front end UM. You know, going back to the heart of active investing,

you've got to have a view on asset prices. Right, so you're trading some universes I don't know, US equities something like that, you've got to have a view stock by stock what's the price is going to be in

a day? Um, uh, two weeks, a month, so on and so forth, right and so UM that front end is called signal generation or generating alpha's right, using data and machine learning techniques to come up with anomalies to that that you identify, and then you build models upon to sort of make a prediction of, um, what the what the what the price is going to be. So there's all sorts of types of data and uh algorithms

that people use. Historically, much of quant investment has been building um what are called um quote unquote technical models, wherein basically you're using historical price and trade data to forecast future price movements. Right, so you might think of things like momentum or reversals or so on and so forth. That's you know, leveraging you know, kind of purely um technical data from the markets. UM. What we've seen emerged really over the past ten years is there's also been

a shift to sort of quote unquote alternative data. Right, so you might look at things like you know, everybody's heard the famous story of satellite images of parking lots, right to try and assess you know, um, is you know, what's the occupancy at Walmart. This you're gonna they're gonna

make their earnings. You know. Quantitative investment would take that kind of data and leverage it to a model which forecasts, Okay, um, what's the return going to be up for Walmart over the next week, the next month, next two months, and so on. Right, So the front end you have um uh this this identifying the data combined with the machine

learning technology which is going to build build predictions. Now, oftentimes you're looking at um or I should say really always these days, you're looking at having many, many um anomalies. So you may have a technical model based on momentum and reversals. You may have bought a whole bunch of parking lot data, you have some some model for the retail sector based on that. You have some some credit card data, some social media data, maybe some some news data.

You have all of these. And so the second part of the process is to kind of uh combine these different types of signals or views into sort of one compositive view because at the end of the day, um, all you care about is is net and that is this asset price is going to go up or go down, and and and that part is called alpha mixing or signal mixing. Right, Um, you have these these separate models that that that you've built, and you want to combine

them to one kind of uh composite view. So um, that's that's kind of the front end again, having a view on what prices are going to be over the other relevant timeframes. Historically that is where the vast majority of U the energy was spent. The idea was that if you have a good signals, if you have good predictions, you can make money. If you don't have good signals, you're not going to make money, and the rest of it doesn't matter so much. So I believe if you

don't have signals, you're not going to make money. That's certainly true. But these days the market has gotten competitive enough and there are enough kind of quant players that, um, what you do with the signals also matters how you try to to monetize them. So here the kind of the next step is that you have Now you know, you're waking up to open to the market. It's nine thirty in the morning, right, you have a prediction for

you know, a universe of three thousand US equities. Now you have to kind of decide, um, what's the target portfolio you want to form? So that's kind of a called a portfolio construction case, right, And so the kind of things you're thinking about are balancing sort of risk versus return. You know, you don't want to be um um uh longer short, maybe you want to be market neutral. You don't want too much exposure in the individual set ters,

you know, UM, so on and so forth. Right, UM, you're balancing that also with with with with transaction costs and so on, and you kind of decide like, um, you know again based on what my current view is of the world, UM, what's the target portfolio I want to hold? And this is something you periodically revisit. It used to be sort of um quants sort of you know, traded once a day and had a trade list the beginning of the day and you know, um generated trades

and revisited the next day. Now it's much more of a continuous procedure because you know, as as the market evolves and as you get more data and news comes out and so on, those underlying views which you're driving the trades are changing. So so that's the kind of the middle piece, UM, figuring out what what portfolio to hold, and and then the final piece is actually um, sort of generating the trades. Sometimes quants do this themselves. I think more and more quants of doing this themselves. Um,

you can farm this also to UH. Basically every major bank or prime broker that services UH quants has an agency algorithms desk that will do this for you. But here here the idea is, Okay, I decided I need to buy two million dollars of Google stock over the uh the next fifteen minutes. Um, how can I do that? Um? You know? Should I use exchanges? Should I use dark pools? Um? How should I uh spread that out over time? Um? You know, should I use limit orders? Market orders? Um?

This kind of thing? And uh again, Historically, UM, you know, people focus a little bit less on that, but now as the market has gotten more more competitive, it's also being important. If if you're not doing those latter two phases, the portfolio construction and the trade optimization, well you're you're leaving money on the table in a way that almost

may may not be may not be profitable. I think one thing that's that's not obvious that or I should say, it's quite different about plant trading versus uh, other types of hedge fund trading. If you look at a guy like um, you know, I don't know, just to sort of pick someone random like Bill Ackman, right, um, when when he goes in and buys a stock, he has like, you know, really kind of strong um conviction. He takes some massive positions, and he also he probably expects to

make or you know, something like that. Again, I don't do that type of trading. I don't know, but he expects to make tens of percents, right, a quant in any individual position, you probably measure your expected profit in basis points, right, And it's all this and you know, you might expect to make three basis points on the transaction cost of two basis points, right, So you really like carefully controlling your costs and managing execution and so

on is extremely important. Like you know, Bill Ackman, if he thinks he's going to make on a particular trade, it doesn't matter if he's paying uh, you know, two basis points for twenty basis points or even a hundred basis points, right, he's going to make so much more in his mind, it's irrelevant. Whereas for for for quants,

you're really operating on a very thin margin. First of all, that was a sort of great explanation of the whole process, really nice overview, But I want to go back to just these sort of search for the original signals or

search for this sort of the the initial inputs. And I'm thinking about large tech companies like Microsoft and Google and Facebook and how they have a lot of like researchers who are engaged in sort of pure tech research and you know, always out there filing patents, and there's probably a long sort of distance between anything that they discover and their own research budget um and then what ultimately might show up in a consumer product or a

business product. And I'm curious if there is sort of an analogy in quant land where you have people who really are sort of at the frontier without a sort of crystal clear idea of okay, this is going to lead to something that will turn into a trade. But it's that process of sort of really exploring that frontier which eventually leads to concrete ideas that do lead to trades.

And I'm curious if that's sort of like the analogy and how investors and how the portfolio managers think about where to explore and where those frontiers are, and where to invest expensive sort of time, energy and computing power in discovering these alpha generating signals. So I think quantity of investors operate quite differently than some of the research groups and in big tech places, like if you go to a place like Google Research or Microsoft Research, it's

really not that different than an academic institution. Um, their their main output is really papers, right in journal papers, conference papers, so on and so forth. Uh, And it's really just a different way to do uh, almost academic research, kind of the classical Bell Labs model. And and maybe I mean they they do consult on internal projects and so forth. But I think in the in the in the quant world, it is much much much more applied.

So I think typically the kind of thing would be like, um, you think, you know, maybe someone comes to you a vendor, or you identify a data set that you might that you think might have some relevance. You start looking at building various models of trying to predict prices or you know, things that are relatively of relevant to prices. You try

and parent some different machine learning kind of techniques. But I think from the beginning it's really oriented around concrete things like let me build a price for let me build a model, sorry for for what the return of this asset is going to be over the next month, right, Or let me build a model for how I should efficiently trade large box of stock over the next fifteen minutes. Broadly speaking, it's much less of them sort of blue sky research that that isn't to say that some people

don't do that. I think, uh, I think people do. But um, the the incentives aren't there because you know, for the most part speaking practitioners, um, there's no publishing, right, and um, I think people are extremely paranoid and sensitive because if if your IP leaks and other people do similar things, maybe what you do will stop working as well.

And so there's, uh, there's not that much of an incentive to have to do that versus the very kind of visceral incentive of you know, making money, having you know, outperforming in the market in the in the short term. So so research in the quant world, for the most part, tends to be much more implied. I have a sort of related question, but why is why is quant investing

or why are quants so um secretive about everything? Or I mean I don't want to call them weird, but there is this sort of like odd culture around quant investing. And you think of places like Renaissance and Citadel, they're all sort of shrouded in mystique. I once heard that Citadel had an original Enigma machine from World War Two in one of its offices. I don't know if that's true, but just the fact that people are saying this kind of thing tells you something about how they regard these big,

storied quant companies. Why is there this very specific culture, mysterious secretive culture. So I think, broadly speaking, um, people in the by side teople in the Hedgeman industry are are generally secretive, but I think the with with regards to sort of their internal I P and and processes. But I think the nature of I P in the quant space creates incentives for people to be more secretive. Right.

So again, just you know, pulling our hypothetical kind of Bill Ackman example, if he identifies some asset that's undervalued, UM, he's going to be sort of very quiet about it until he goes in and accumulates the position you watch, because he doesn't want other people to know and other people to front run him and to sort of take

that opportunity away. Now, once he's a mass, that position perhaps will actually start even advertising it, right because now if sort of people sort of follow him, works to his benefit and it will push prices in the way that he wants. The quant space doesn't quite work like that. Like again, any individual trade is a very short rise and maybe a couple of weeks. Right. Trades are sort

of very small and diffused across many many assets. But the idea of the trade, the data source coupled with whatever is generating the signal and the uilarity methodology and so on, that has lasting value that might you know, work for for the next six years. Again, year on year. It will the performance goes down as anomalies disappear, but uh, you know, it has multiple years of value. So uh.

The general feeling is if people sort of figure out what you're doing, and um, where the opportunities are, and what data sets you're doing and so on, they will also do a similar kind of thing. That will they will copy you and then those anomalies will disappear faster, you know, at least in my experience, because of the longer time horizons over which this um, this IP decays

people are more paranoid about being extremely secretive. And that's not only for for outsiders, but that's even within firms.

So so many firms are siloed down to the level of individual quant researchers, where um you maybe um uh you know, you may have a team of a couple dozen people all um uh let's say under a single PM, all working on the same overall strategy, but you won't know what the guy next to you is working on, right, And if you pass data sets across maybe you um label them in sort of random ways and so on. So no, nobody, nobody, sort of maybe has the full picture except a handful of people um on the top.

And again the idea there is that you know, over time people quit or leave or whatever, um you want, the firms would like them to have as little of the I P as possible in terms of uh, you know, not decaying the value of their own IP. Now, I think famously renaissance does not operate this way, So ret renaissance is um. One example I've heard where a firm which is uh I think, very very difficult to get into in terms of being hired. But but once you're

in there. They're quite open in terms of what are the different things we've tried, Where are the things that are working now, where the things that haven't worked before, and you know, so on and so forth. And I think actually from the perspective of research, that works much better um quant researchers tend to, uh believe it or not, tend to be kind of social animals, and it's it's always more fun to work on things with other people rather than just sort of sit at your desk with

with the with the blinders on and so on. You know. Interesting about Renaissance is how they've been able to manage it so that very very few people have have have left and it seems like, you know, they have not had the kind of ip loss that other people worry about. So Renaissance famously just puts up extraordinary numbers year after

year after year. And the sort of the trick or one trick besides there being a bunch of mathematical geniuses, is a having this sort of open culture of collaboration and research and be somehow preventing a lot of exodus so that no one else has really been able to replicate their approaches. Uh In any way, how hard is this? So you think about like someone like I don't know, like you hear about other other managers, like you know, Steve Cohen is like, oh, I wanna allocate money to

quant How hard is it? And this is sort of something I want to explore more now, is like, how hard is it to sort of anti up into that game and to sort of start being competitive if this if you're sort of starting from zero right now, Um, I think it's uh, it's a tough place. It's a it's a it's a competitive game. Maybe not so much anymore. But over the past five seven years, Um, my general perspective is that the buy side active managing sort of

hedge funds have been shrinking overall. The one sector that has not been shrinking his quant and so I think there has been an entrance of uh kind of new players there. Um. Now, I'm Steve Cohen you specifically mentioned. He's actually been at it for a while. He's been in the quant space for the since the early two thousand's.

He on the order of twenty to thirty percent of his assets are actually quant some some something like that, like a non you know, people mainly think of him as a long short uh kind of guy, and that's probably mainly what he is. But again, um, you know, maybe a third of his assets are in a quant space through after Cubist and so on. Now he operates

very differently, um, he operates. His quant funds operate in uh kind of like traditional UM long short guys operate, wherein you hire individual pms, you watch them a very kind of carefully. They make money or they lose money if if they if they're not making money quickly enough, you fire them. And uh you sort of you kind of have a portfolio of these these individual managers who are who are doing their own things, who are tightly siloed, and uh you know, uh you try to manage that.

And that's the way his quant operation manages. So there's you know, again, um, a whole bunch of small um let's call them pods or whatever, of you know, two or three people each kind of doing their own thing in an uncoordinated way. You know. That's again quite a different model than let's say, the Renaissance, which is uh um uh you know, one kind of open strategy. And I think the the advantage of the Steve Cohen model is that uh Um, you know, uh, it's it's easy

to to hire people from the HR. Process is very easy. You don't have to care when people come and go on so on, because you're not really investing in any of their their individual I P. Right when when someone leaves or like let's you fire someone, it's because they didn't do well and whatever they have is maybe not worth um that much and they don't know anything else about what your other p are doing and so so

that process is very easy. But I think the downside is that what we're sort of starting to see is throughout the quant space, like you know, the broader technology industry, we're starting to see that there are a lot of increasing returns to scale that as you get bigger and bigger firms are able to build advantages. And one kind of concrete source of this is around trading costs. Right, Um, when you're thinking about, like let's say, on an individual trade by trade basis, do I want to get into

this trade? You have a prediction of how much you're going to make if your if your models are correct, but also there are these costs that you're paying, these transaction costs, and if your prediction doesn't exceed your costs. You shouldn't put on that trade because even in the

best case, you're you're not going to make your money, right. So, um, what what's happened is that as more and more people have gotten into the quant space and more and more of these identically anomaly sorry are identified and markets get the more efficient, the signals have gotten weaker, right, and so um, just to sort of give a give a maybe a concrete example. One signal that's sort of um, quite well known throughout the klant industry and un academics

of published papers and so on, its order book of balance. Right, if you go out and you look at an electronic order book and they're more buyers than there are sellers in terms of the resting limit orders. Um, it's it's more likely that the press will go up and go down. Right. You can. You can go out and try that that

has a predictive value. Now, however, if that's all you know, you won't make money because you might think the price is going to go up you know, a tenth of a basis point just to throughout a number, but your transaction costs or two basis points, and you know you're just not. Um, you can't exceed your your costs. So, um, the transaction costs to a first approximation, they're they're kind of like on a trade by trade basis, a fixed

costs that you have to exceed. Now, if you're in a world where you have many, many signals, maybe tens, maybe hundreds, maybe thousands, and you're adding them up and they're independent and you trade when they're all aligned, now you can have sort of, um, you know, signals that are weak individually, and nevertheless, when you you combine them, when you aggregate them, you are able to exceed transaction costs and monetize them. So that that order and balanced

signal that I just sort of talked about. If you're sort of one guy in your basement and that's all you knew, you can't make money off that. But if you have twenty other signals and you're you know, you're going to put on a trade anyway, in some sense, the transaction costs become a sunk cost, and and that you know, point one basis point that you're going to get because of this well known signal, that becomes free money. So so as you get that kind of economies of

scale because of fixed costs. I think it becomes harder and harder to have UM quant strategies where um, you don't have a lot of people UM you know, in a very kind of coordinated research process where you have people working essentially independently UM the kinds of UM you know places that are structured like like like let's say Renaissance again, where you might have like two plant researchers all working on different aspects of the thing, and then

you know, these things combined to one sort of overall view of the market. I think that is able to sort of better monetize a lot of these signals in

this kind of more competitive world. So on that note, if if you are running a lot of these strategies, getting a lot of these signals, and you're able to lower your transaction costs because of that scale, and at the same time, quant investing has these big barriers to entry because you have to have these technological outlays, you have to hire a bunch of PhDs and things like that. Does that mean that the industry is inevitably sort of

trending towards a monopoly? Are are we going to get a situation where there is just one or maybe two or three really big quant investors because no one else can compete with them effectively. I think we're kind of there. I mean, I think there are only a handful of large quants. Most of them have been doing it for a long time. I mean, Renaissance, d SHAW, PDT, two Sigma. You know, there's there, there's a there's a handful of others.

I think it's it's harder to see, you know, maybe there's some exceptions, you know, in terms of funds that have launched UHM more more recently, but it's difficult to see people of of that scale with with the similar track records. So I think we are seeing some degree of consolidation. I don't know what the altar I mean, I don't know if it's gonna um come down to

one firm. I think, you know, probably not, just probably kind of more competition, but I think it will be harder to have uh sort of either either more independent managers or like UM kind of the siloed model of places like uh you know, UM S, A C and Millennium. If I want to start a quant fund, what are we talking about in terms of how much it's just gonna cost for computers and data just to give even get in the game. Don't do it, Joe. I feel

like this whole conversation is how you shouldn't be doing that. No, I realized, I realized that it's a bad idea. But let's say I'm an idiot and I try anyway, Like, what are we talking about? So? Um, I think things have gotten over time much more expensive, things like data feeds and uh you know, so on the exchanges have

constantly been ramping the prices on on these things. But um you know, these days what's become one of the biggest costs is actually just pure computation and and this is also a trend we see um uh, you know, more broadly in a technology, you know, if you look at kind of the state of the art models for things like um uh, computer vision, object recognition for um uh, you know, playing games like a chess and go and so on, these types of models leverage approaches and machine

learning that are really based on having a lot of data and doing even more than that, doing a lot of computation and and and so the spirit there, you know, coming out of places like deep minded Google or be ai and stuff, um open ai, um of you know are artificial intelligence UM company that their main model is is literally like we're going to do simple things, but we're going to leverage it to massive scale computation, right, and so so I think you're starting to see that

in finance as well, where you need to do things like let's say you need to um UM back test a trading strategy. UM, but you have some parameters, and you want to try tens of thousands of combinations of those trading parameters, and each one involves a simulation over you know, twenty years and so on and so forth.

You need a lot of computers. So UM. Someone told me anecdotally that at a major quant shop, each quantitative researcher has given kind of a quote unquote budget of of of ten thousand CPUs, right, so I ain't given time, they can use up to ten thousand individual kind of processing units. And just to give you a sense of what that costs, UM, you know, if you're to go, you know, buy that on Amazon at AWS, that would

be the order of magnitude maybe a million dollars a year. Right, And this is just for this is just for research. This is not to actually generate the trades or whatever. This is just a tune all the parameters and and and sort of really optimize your performance. That's really interesting. It kind of makes me wonder how how good I guess academic research is at gauging quant strategies if the outlays just to run a few experiments are so massive.

But on a slightly different topic, I wanted to ask you, I guess this question is kind of inevitable. Whenever you talk about algorithmic trading or systematic trading, what value do you think quant investing actually creates for society? So, for instance, when we talk about traditional investing, that's supposed to channel capital in the most efficient way possible to good companies, and that should in theory benefit the entire economy. But

quant investing, as we've discussed, isn't really about that. It's about arbitraging these small differences. So maybe it makes prices slightly more efficient, but is that worth the enormous infrastructure investment that we've been discussing being spent on it? So I think there is, um there are some benefits. You know, it varies based on the strategy and based on really the incident time. But I think a lot of uh,

you know, to to a first approximation. If you see a price move in a direction that's unusual, UM, it could continue or it could revert. Right to the extent that you think it's going to revert, you're going to sort of bet against it. And what what that amounts to is basically supplying temporary liquidity to the market. Right, So, I think the positive aspect to UM quantitative investing is that UM I think a lot of it is supplying liquidity to the market on a horizon of uh, let's

say days to two weeks right now. The flip side is if you're if you're really it's more of a momentum that you might be accelerating the trends UM you're taking away like wuity, you're competing for that liquidity, but as you said, maybe you're making prices more more efficient. So I think, on balance, I think that that probably

there is some benefit. I think it's probably small. Admittedly, is it worth all these uh, you know, very smart people being drawn away from other fields and so on, I'm not sure, But you know, probably as much or more resources or spent at places like Facebook and Google getting people to click on ads. Right, I'm not sure

that that's uh as positive the pressing. Think about all these people, um, you know, looking for signals to squeeze out three basis points in the market, because there could be some great innovations in squeezing more ads onto a mobile phone that they'd be working on. And there you go, kind of a sad allocation of resources. See you think Joe's joking, but he's he probably doesn't. So uh, here's

one thing that also always tends to come up. It's this idea of um, this type of trading reaching the limits of available technology and pushing the strategies to sort of greater extremes. But those extremes eventually have limits. And so I guess I'm just wondering, is there a limit to quant investing? Is there a point at which quants sort of arbitrage everything out of the market and the signals are no longer useful or the algorithms themselves are

impacting the market in some way? And on that note, what's what's the next big thing in quant investing? I guess yeah, So, I mean I think there's a constant balance. These are finishing of inefficiencies are being identified in arbitrage the way because there's money in it, right, and so as arbitraged in it, Um, the money sort of disappears and then you get sort of a fewer people kind

of doing it. But um, so long as there's uh, you know, kind of traders out there, we're not paying attention to this stuff and uh you know, the Robin Hood traders or whatever and are kind of leaving money on the table, Um, there will be people um there who are trying to uh sweep up the crumbs in terms of where it's going. What the what the next

big thing is. I think it's it's it's pretty hard to predict, but I think um, uh you know, broadly a shift towards uh things that are even more black box, even more computationally driven and uh um not so much. Uh you know, have like kind of nice structural explanations. Um again sort of following a lot of what's going on in in the tech world as we shift to ideas like deep neural networks and reinforcement learning and so

on and so forth. Um. You know, you know, again, you have these these systems that work worked great for Let's say I'm playing go, but it's really hard to explain what's going on and I think we're starting to see that in the quant world as well, again leveraging a computation but really really ending up with with things that are you know, um, you know black boxes that

you know just are completely not transparent. So in other words, you know, like you could look at something like satellite images and say, oh, there's a lot of cars parked of Walmart and then predict the Walmart stock is going to be up. But the next, um, the next generation of things to watch out for is this works, and it works consistently, but we as humans can't really articulate

why exactly. That's super interesting. Well, on that note of humans not really even being being able to explain what they're doing, um, it seems like a perfect place to stop. Thank you so much for joining us. Thank you so much, Tracy. You know, uh, as a as a media person, I have my own experience with the sort of alpha decay that CMX was talking about. Do you know what it is? Um, did you build some sort of algorithm to take advantage

of like Google Ads or something and then it stopped working? No, there's nothing so sophisticated. But back in the early days of like blogging and stuff, I remember this phenomenon where you would come up with some like headline construction. You'd like five things you need to know today. Remember, like the old upworthy headlines, they were like and you could

and you can't guess what you know. And then those work and those generate like excess traffic, and they get shared on Facebook, and then everybody discovers that these headlines cliches work, and then everyone does them, and then people stopped clicking on them, and you need to like find

I don't do clickbait anymore. But I always thought at the time like that was like a very similar process to uh, to this sort of quant approach to investing, the sort of search for alpha and alpha decay of a blog headline anymore was the key word in that sentence about clickbait. But I think it's a really good analogy. It is a good analogy because like the usefulness of those headline constructions decays over time, as you point out,

because more people are copying them. But it also kind of gets to that point about the limits of this type of investing. There are only so many ways that you can construct a headline, and eventually people kind of catch on two different ones and they become not so enticing, and I kind of wonder if the same thing could

eventually happen to quant investing. So obviously there are many many more possibilities in quant investing, and it's possible that markets are always changing and so opportunities for arbitrage and identifying these signals are always coming up. But it does make you wonder. It certainly does. And what he's talking about at the end, where maybe the signals of the future are just things that work it can't be articulated, is just like a super kind of fascinating phenomenon to

just like wrap your head around. Yeah, I feel like that's a good microcosm for maybe the human experience in the future. Like we have the technology, we're not entirely sure how it works, but we're just going to sort of let it run and hope for the best. One of the things that sort of interested me is like a sort of thing to watch going forward, is okay, So we talked about a huge aspect of that was just the costs and how like you might be able

to identify a profitable anomaly. But unless the cost of getting the data and executing the trade is lower than that, it's um, it's useless, but you know, you also have to wonder, like, okay, right now, like a certain handful of exchanges, say, control a lot of the trade data costs. In theory, that seems like an area where maybe new entities will come and find a way to provide data

cheaper Amazon Web services. You know, presumably computation costs are going to keep coming down, and obviously that was a big breakthrough from probably the old days where you had some sort of main frame on premise services. You know, computation has gotten cheaper, so there's probably always going to be new opportunities to squeeze out even smaller profits because there are ways to shave costs in sort of your

in your research, your work. Yeah. Maybe the other thing that was really interesting was the idea that quants um. I think CMAC described them as actually social animals, which kind of flies in the face and I think of a lot of stereotypes. But I'm also I'm really curious. I would love to be embedded in a firm like Citadel and just observe how they work together and what's considered a good out go, a good systematic strategy versus

a bad systematic strategy. Obviously you wanted to make money, but are there certain things that are more valued over others? Maybe cheapness to execute or um, I don't know, risk management, something like that. I'd be so curious to see how that all works. Um, I'm sure if we just walked in there, just let us in the door and let we could just hang out there for a while. Yeah, I'm sure they wouldn't mind at all. No, let us see their white boards stuff like that. Citadel, if you're listening,

we would like to come sit you. Okay, should we leave it there? Let's leave it there. This has been another episode of the ad Thoughts podcast. I'm Tracy Alloway. You can follow me on Twitter at Tracy Alloway and I'm Joe Wisntal. You can follow me on Twitter at a Stalwart. And you should follow our guest on Twitter cmx Millenmy he's at Cmax. Follow our producer on Twitter, Laura Carlson at Laura M. Carlton. Follow the Bloomberg head of podcast, Francesca Levi at Francesca Today, and check out

all of our podcasts under the handle AD Podcasts. Thanks for listening to the year