We've talked about many of the ways AI is working itself into our lives at the office in Hollywood, even in the US presidential campaign today. Another place where robots are moving in Wall Street. Of course, traders have long used data to try to divine which way markets will move. Quantitative traders are quants already build sophisticated mathematical models and amass huge amounts of information to identify opportunities others have missed. But AI can conceivably do it even better by pulling
in and analyzing vastly more data and much faster. But is AI better at beating the market than human stock pickers. Bloomberg's Justina Lee and Sam Potter say so far, the results are mixed.
Part of the issue is when you're using machine learning algorithm, you lose the explainability of it. So when you're losing money, your clients might be like, why did you decide to trade that?
I wonder if someone will discover an achilles heel to these machines in the way the machines work that will provide an unintended market advantage for those who are looking for it.
Justina and Sam are here to tell us how investment firms are using AI and how it actually works, And later we speak to Renee Yao, whose firm is leveraging this new technology.
We're not just saying we're using AI to generate supure returns. We're using AI combined with good risk controls.
I'm wes Kasova today on the big take. Can AI supercharge your returns? Justina, We're hearing all the time about how AI is encroaching on every little aspect of our lives. If I suppose it was only a matter of time before the markets wanted to use AI to make money, and you found there are so many big brains working on this question.
Yeah, I mean maybe we can start with a certain category of big brains on Wall Street, which are these people we call quants. What quants stand for is quantitative traders or strategies or analysts, and they're basically people who are using computer models to figure out how to make money in the markets. Like usually it's about analyzing some
pattern and trying to profit from that. And in recent years, quants have been especially interested in using a certain kind of AI called machine learning, which basically just means that they're using very sophisticated techniques, you know, the kind that
is powering chat GPT, that is powering driverless cars. They're using those kinds of techniques to try to figure out exactly, you know, where the S and P five hundred is going, or where the Apple Stock is going, and trying to make money from that.
And Sam, that's a really good point because Wall Street has been using computers and modeling in a really sophisticated way long before any of us were talking about AI and chat gyput.
That's absolutely right. Whereas and I think one of the things that Justina and I talked about through the process of this story was when you look back over the past ten years or so, especially places like Bloomberg News and other platforms have repeatedly said, you know, AI is coming further jobs on Wall Street, Computers are going to replace traders. You know, humans are going to be a thing.
Of the past.
The AI revolution is upon us. We see those stories time and time again. But one question that we had through this story was where is it then, especially in a year when chat GPT has blown up, everyone's in a frenzy about it. It's powered a lot of the stock gains this year has been this excitement over stocks like Navidia who are building AI technology, and yet wall streets and the business of investing seems no closer, Like
AI has not displace the humans yet. Yes, quants are out there and they're using the latest tech, but where's the revolution? Is it still happening or are the human trade as say for a while?
Yeah, Justina, what is the difference between what I guess you would call traditional quants, that people who are digging into numbers and making models what they do, and what AI platforms are now trying to do.
Yeah, that really is a great question, and it's mostly a difference in statistical technique. So for traditional quants, a lot of the time, if we simplify it, what they're trying to figure out is the relationship between stock returns and some variable. Right, So for instance, you might look at Apple's profit margins and decide that it's a great buy.
So quants are doing that, but they're doing it statistically, and to simplify a little bit, usually they're using what we call linear models, which means they're saying this leads to this, which is a bit simplistic, but at least you can understand it and you can kind of understand why that does not always work out because there are so many variables moving a stock performance every single day, and what machine learning algos generally try to do is
it can take in hundreds of variables, all the information you can find, and try to figure out what the relationship between all of them is and how they predict the price performance. The downside of that is that you don't really understand why the machine has decided to tell you to buy a stock that day. But the flip side of that is that it kind of comes closer to just how complicated the world is.
And so SAM is part of the idea that the models that quants use are based on a certain number of these variables that they put together to try to make predictions. But these AI models are able to take in many, many, many more of these and paint a more sophisticated picture of what's happening in the market.
I would say that YESAI is taking into account many more variables than the traditional quant methods. Traditional quants, many of them, are trying to adopt these adapt these layer them onto the theories, the academic theories that they built all those years ago, and there's kind of debates within the industry over how effective it is to do that.
And you know, we always hear that the world is producing so much more data these days. Quants are also looking at our tweets, for instance, or our news headlines. I mean they're not actually reading them directly, but they are using a kind of machine learning algorithm called natural language processing, where they try to turn all the material in the world into numbers that they can analyze.
And I guess that's really sort of the holy grail of this right is trying to understand the human behavior that underlies all the decisions that go into markets moving up or down.
Yeah, exactly. I mean that's sort of hardest part. I mean, if you can predict where the S and P five hundred goes every day, I mean, you would be a rich person. And that is the hardest part, especially in the US thought market, for instance, which is so efficient. So even in my story, when I spoke to a lot of these quants who are so optimistic about the progress in AI, they always say, you know, we're just looking for a small edge, But a small edge on
Wall Street can mean a lot more money. So even if you're not right every day, if you're right, more than half the time. I mean, that's good enough.
Justin in your story, you spoke to a lot of the people who are doing this work. Are trying to incorporate AI into trading. Who are they?
Yeah, obviously a lot of quants will say they use machine learning, but there are some hedge funds that are known for the AI prowess. So, for instance, I spoke to Michael Kurrichanoff, who's the co founder of Volon, which is one of the few hedge funds known for AI. And he's a guy, you know who was a physicist, you know, who has a computer science PhD. After a period in working in Silicon Valley, he went to Wall Street.
He started bolly On about sixteen years ago with the idea of using machine learning to try to predict where the equity market will go. And if you talk to people like him, one interesting thing is they almost sometimes sound a little bit scornful of all other quants, in that they really see themselves as doing something that's different from the last generation of quants, and their pitch to investors is we can give you something that doesn't look
like anything else out there. So even when your other quant strategies fail, we will be making money and.
Sam exactly what is it that they're telling potential clients that they're going to be able to do with this technology.
I think one interesting thing about Boleion is that the perfect example of this run by a data scientist who used to do nuclear physics, and they based themselves near the Berkeley campus so that they're close to a center of world machine learning. The people trying to harness this AI and beat Wall Street and not your classic finance types.
They are data people, they are statisticians, they are programmers, and I think that is an interesting dynamic that this sort of power based on Wall Street is potentially moving away from the classic finance economist types and towards mathematicians. Basically, certainly for Volion, they're presenting something that is uncorrelated. You imagine you're a giant investment firm and you're placing your money just directly in the stock market or in a
track of fund and in some bonds. You basically create a risk exposure that is the same as the market. You are vulnerable to the same things. So a lot of what people are looking for on Wall Street is uncorrelated. It's why we see a lot of activity in occasionally very random commodities funds and things like that. Because these big investment houses are trying to spread out their risk,
they're trying to be smart about it. Cracked, beating the market, beating the benchmark doesn't seem so just yet, they do say they are providing a return that is going to behave differently to anything else, and that's very appealing to certain money managers.
So is the idea that when the herd of the market is zigging, that AI will allow these firms to zag and take advantage of opportunities that the rest of the market just doesn't see.
Yeah, I think that's exactly right. And the reasoning behind that is probably that they are taking in so many variables from everything that they can get into a more sophisticated, superhuman view of what markets are really doing. And one thing that they really emphasized is they don't have a model. I mean, they don't have an economic theory, but they just trust the data. You know, in the sense that you know, chat GPT is reading so much Shakespeare that
it can write Shakespeare. So it's the same idea, which is that the machines are looking at so many years of market data that they can tell you based on what the market looks like right now, here's where prices are going to go. And even if you cannot reason it as a human, it will tell you the right answer after.
The break, are these AI models beating the market?
Same?
I guess on the flip side, taking advantage of all this information may not necessarily lead to a clearer picture. We always try to figure out why the stock market went up or down on a given day, but often it's not really clear. So does all this data that says what happens in the past necessarily predict what's happening now.
I've got a smile on my face because you're talking to a guy who ran Bloomberg's Global Wrap for about three years. So my job every day was to define a narrative. You know, why why stocks are up, why bonds are down, et cetera. And it's true. And this is where the kind of rubber meets the road in the story, the AI and the machines are finding it just as tough as everyone else to figure out what's
the driver and where it's going next. One of the problems is regime shifts happen in markets from nowhere when the pandemic hit, we obviously had a massive crash. For anyone to see that coming machine or otherwise was near impossible. And then afterwards, with the various pandemic support and emergency aid from governments, we got a rapid rebound and some of the AI funds were caught out by that. Basically, the AI the machine. It can look at the complete
history of all stop moves. Ever, something can happen, something external, something macro, that makes tomorrow different, tomorrow a thing that never happened before, and that's probably there. Arguably the biggest shortcoming is you can learn all you want from history, but it doesn't tell you what's going to happen next.
Justine, you also talked about another fun called Man Group. Yeah.
So Man Group is the biggest listed hatch fund in the world and they're a huge company based in London. And what I found interesting about their story is they actually started looking into applying machine learning back in two thousand and nine, but they didn't really start using it
and actual trading until twenty fourteen. And that just kind of tells you how much research and work goes into getting it right and making sure it actually helps and how even though we've seen so many breakthroughs in Silicon Valley, the financial industry can feel like it's many years behind. Part of the issue is when you're using machine learning algorithm,
you lose the explainability of it. So when you're losing money, your clients might be like, why did you decide to tred that you can't exactly point to a formula or point to your reasoning because machine learning, I mean, the beauty of it is that it can figure out all these complex relationships that you cannot exactly reverse engineer rationally, and so it's really hard to explain and that is a bigger issue in finance than it might be. You know, for chat GPT for instance.
Same is there any downside to putting so much into these AI models? When you look at something like chat GPT, one of the big problems is what they call the hallucination rate, where it just gets it completely wrong and yet asserts it with total confidence. How do we know that these models are getting it right? We don't.
And that's where the real debate in the field is coming from those who say we need to overlay some economic theory so that they know they have some idea what they're looking for. What the advocates of just turning the machines loose would tell you is that if you let them loose on the data, they will find the old school quant rules at the same time that they find everything else.
And I think that's also why a lot of hatch funds and financial firms aren't using it to predict prices and trade based on that just yet, because they really want to be sure that's better than everything they've ever done.
After hearing this and seeing how so many firms and smart people are in are being put into this question, is it working? Do these AI models beat the market? Do they do better than the old fashioned and new fashioned ways of trading?
I really want to josh the question and.
Say the cury is still out.
I mean, if you look at the track record, the answer definitely is not mind blowing. So in the story, I cite this academic paper that looked at mutual funds that use AI and what have found was that most of them still did not beat their benchmark, even if they were a bit better than the human managed funds. And in the exchange traded fund market in the US that pick stocks based on an AI that has a pretty long track record, and that also has not done
that well. And I think if you look at the more sophisticated AI hatch funds, what people would say is, you know, they've done pretty well, but it's not going to be like number one on the leaderboard and it's not going to like blow your mind. What was clear to me is that the use of these methods is increasing, but a lot of the time they're using it but not necessarily letting it go wild just yet.
What we've seen with quantum besters and quant strategies of the old school is that when they found something that worked, everyone quickly piled into the trade, and then suddenly it didn't work anymore because there are too many people doing it. The gap gets closed very quickly and they can't make money. As these AI programs get more sophisticated, more embedded in the market, any inefficiencies they find, they're going to disappear faster.
The market actually could end up way more efficient and making excess profit, excess return. Beating other people in the market might just get harder and harder because the computers get so good at it.
So we know that this technology looks promising, isn't quite beating the market yet overall, But when you look down the road, given how quickly this technology in AI in general is advancing, what do you see, where do things go from here? In how it's used in markets.
In terms of actually using machine learning to predict prices and decide what to trade. I think that's probably going to happen, but at a slower pace, just because it's much harder to make that decision to hand the reins of your money management to a machine that you don't understand that well.
Sam Justina, thanks so much for sharing your reporting. I have a feeling we'll be talking about this again.
Thank you so much for having us.
When we come back, we'll talk with a trader who's all in on AI. Now, let's hear from someone who's actually putting these AI models to work. Renee Yao is the founder of neo Ivy Capital. It's a quantitative hedge fund that uses artificial intelligence in its investment decisions. You are a quantitative trader. Exactly what is that and how does it differ from AI that's not being used.
So from a normal person's standpoint of view, market might be full of noise or chaos, but behind those chaos or randomness market actually does appear certain patterns, and AI works better as a noise cancelor compared to traditional mesters. If you look at the history of QUANTU investment, it
actually undergone three different generations of big changes. The first generation quants probably dates back to nineteen eighties nineties, where people like Solomon Brothers are trying to mimic what the financial analysts we're trying to do and they use simple implementation methods like Excel spreadsheet to do this their calculations.
And later the second generation quant they realize they not only need to have a forecast of each of the stocks, they need to have different forecasts that has uniqueness that others don't have, which is why they're trying to hire hundreds of thousands of people to try to achieve that unique edge.
And they're all trying to create computer models and other methods to look at the market and find holes in the market that other people don't see.
Yes, they're trying to do data minings, pattern recognitions from historic data and trying to see if there's certainly pattern that happened in the historic data and hopefully that's going to happen again in the future. And then what happens is we do we consider ourselves to be the third generation quant which is we're not trying to do data
mining pattern recognition. Instead, we use over AI I augorism to trying to come up with fast new ideas automatically for us, and they look at leading indicators to predict the future instead of the lagging indicators of what happened already in the past.
So you no longer need the hundreds or thousands of individual incredibly smart people all working away on computers to find patterns. You just let AI do it for you.
Yes, in some scenario, I think our AI allg has replaced the functionality of the human researchers.
That's really interesting where you put it that there's o this noise. What's an example of noise? And how does it get in the way of figuring out which way a market is going to move?
So solving the market direction is like try to solve a crime see problem, Like we're trying to figure out who is a suspect. If we have a video showing what happen happen exactly at the crime scene, that will be like perfect signals because it can leads us to directly who the suspect is. However, in most of the time that's not the case. We don't have a video
capture what happened back then. So what we can do is we look at things that are relevant that will give us certain indications as to what the suspect looks like. For example, certain things might lead to hey, the expect late is a male or female, and so AI it's good at capturing rules, little small pieces and put them together, and.
So AI is able to sort through all the different things that might be causing a market to move, and it's always difficult to tell which one of the many many things it might be in zero in on the ones that might be the most important. Yes, exactly, and how does it do that?
A good example of traditional how machine learning organism will be IBM's The Blue All Go, where IBM built and released in the nineteen nineties and they use that to be the Russian chess champion. Now, the reason we can use traditional machine learning for chess game is because chess game has so many well defined rules, like queens can only move in a certain way in knights can move in a certain way, So after each movement you have
only finite possible solutions for next step. That's why, as long as you have enough computing power, you can literally tell your machine to go through every possible scenarios of the next steps and handpick you the highest winning hand. However, we don't have that luxury in real market problem right, Like we don't know where sp is going to land by the year end.
And there's a million rules and so many participants, and so it's not a finite set of variables. Precisely, how effective has it been? Are you finding that these models are much better than previous models without AI?
I definitely think so. If you only look at lagging nicular historic information and then you hope the market will be in a history repeat itself pattern, then that's probably going to have a lot of constraints because we live in a day where every day like something unexpected happened, Like, for example, just the past five years, we experienced COVID, we experienced the sudden collapse of Silicon Valley Bank, one
of the largest regional banks in the US. Those are all unprecedented events that didn't happen historically, So naturally that's going to post a challenge for the traditional much learning models. Where relies on history repeat itself. But for AI, because the model is learning in real time like we do, then it's able to navigate those different market conditions smartly and deliver good returns.
You're not saying, though, that AI has the ability to predict, say the collapse of a Silicon valley bank, but that it's able to respond more quickly to figure out what to do in the event that that happens.
Yes, our goal is to respond quickly and smartly as to what to react if such unprecedented things happen.
So you said you're having good success with this, but if you look at the overall numbers, I guess that these AI algorithms aren't yet doing a great job of just beating the overall market, doing better than say the S and P five hundred.
That's actually one of the myths with AI. People's attitude towards AI tends to go to the two extremes. Some people just don't believe it. They think it's just a ring name of a traditional technology that has been existing for more than fifteen years. Or some people will say, oh AI is God, it can do anything, it can do everything. Well, my experience is AI is not stupid
nor it's god. AI does has its limitations because there are periods of time where markets are completely randomness and there's just no pattern for you to capture, SOI no traditional machine learning will work in those periods of time. However, during the period of time when market does have a pattern, AI works better as a noise canceler compared to traditional mation learning.
Where do you see all of this heading, do you think that AI is going to more and more work its way into markets and work its way into the way trading is done.
I definitely think so. Eight years ago, when I first came out and becoming a pat then build NEYV not many people are interested in muchine learning of AI. But now with Tauchipt, with Task last Auto Drive, more and more people began to realize how powerful AI is and how much more it can be done without just involved over human So that's why I believe in the future AI is definitely going to play a much more important role than today in terms of its applications in the financial market.
Renee, thanks so much for giving us a look inside what quants do.
Thank you so much.
Wes, thanks for listening to us here at the Big Take it's a daily podcast from Bloomberg and iHeart Radio. For more shows from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, or wherever you listen. And we'd love to hear from you. Email us questions or comments to Big Take at Bloomberg dot net. The supervising producer of The Big Take is Vicky Bergolina. Our senior producer is Catherine Fink. Our producers are Mow Barrow and Michael Falero. Kilde Garcia is our engineer.
Our original music was composed by Leosidrin. I'm Westkasova. We'll be back tomorrow with another Big tag