Bloomberg Audio Studios, Podcasts, Radio News.
Hello and welcome to another episode of The Odd Laws podcast. I'm Jill Wisenthal, normally joined by my co host Tracy Alloway, but she's on vacation today, so it's just me in this intro. But in today's episode you will hear a conversation taped live at Bloomberg's Reimagining Information form. On June twelfth, we spoke with Gappi Pallioligo, global head of quantitative Research at Ballysny Asset Management. He has a new book out,
it's called The Elements of Quantitative Investing. Neither of us have read it because it would go way over our heads because we're not quant so we don't know how to read that stuff. But Gappy is great at explaining all of this stuff in clear English. So we had a great conversation and we hope you enjoy listening to it.
So just to begin, I'm going to start with a really really dumb question, possibly, but isn't all investing quant investing nowadays? I mean every investor has access to some form of quantitative or using numbers.
Yeah, I guess yes. End of answer. Yeah, I think so, I think so I mean pretty much everybody uses some kind of quantitative overlay, right, but two different degrees. So I have a friend who worked for one of the Tiger cubs, and they they refused to use sharp They refused to use logs in a spreadsheet because they said that they were dangerous. Probably they took the log of a negative number. And so yeah, no, two different degrees. But yes, there is some quantitative culture seeping through.
Okay, so what defines quantitative investing? How would you differentiate that from I don't know, value investing, discretionary investing.
Okay, I think that there are several possible answers. So I'm going to go with the one answer that I read in my life as a quant I think it's a wily book. It's a very good book, by the way, And I think Cliff Asnes defined quantitative investing as basically investing in a large cross section of assets, having a relatively low edge low expected return in all of them.
And so that's its definition, but it's not quite I think complete enough at this point, because you can also be a quantitative investor trading a relatively narrow cross section of assets but with high high frequency. Right, So What matters really is the number of bets in a sense
that you are going to take right. So I think that probably is if you have a large number of independent bets or quasi independent bets, this means that you need to be able to scale your method to a large number of independent bets, and this means that you are in some way a quantitative investor.
Speaking of roles and jobs, what do you Global head of quantitative research at Pelisney. What's your job? You've been there about six months. What does the job intel at a at a fund, at a firm like PALSNT.
Okay, global head of quantitative research. Okay, So basically I am the head of quantitative research for equities, and maybe one day in the future I will do you know, some commodities or fixed income, but I'm perfectly happy to serve equities, you know, both discretionary and systematic. What we do is I mean my group mostly, I mean I am in meetings, so I don't do any work. So we in a sense provide centralized quantitative services for the firm.
So the first backbone thing that we do is you develop factor models wherever you can, right, so for equities at different horizons. Ideally you would like to develop them for other asset classes. But you know, factor models are the backbone of a lot of quantity investing nowadays. And then hedging at the firm level and at the individual PM levels, which is apparently very simple, but actually it's very deep as a problem. And then we do portfolio
advisory services, which is basically you go two pms. You help them construct better portfolios, You help them understand their performance, which is extremely important, manage their risk, manage their drawdown on occasion, be their therapists. But this is what we do.
I know you're in meetings all day, but you know, if you were someone on your team, how would you be coming up with actual ideas for factors. I hear people who sometimes come up with ideas for all thoughts episodes. Some of them have even turned out reasonably okay, But how does idea generation work? You sit down, You're like, I need to come up with a new factor today. What are you doing? What are you looking at?
Okay? I want to specify a little bit more what's a factor because otherwise gets a little bit too vague. So like, there are factors and factors, So there are some factors that are real factors and what are those? Those are essentially attributes of some kind that you can assign to your investable universe. And there are sources of returns that affect the individual securities through this characteristic, and
they are pervasive. So every asset is in some form affected by the systematic source of return number one, So they've got to be pervasive. The second thing is they got to be persistent, right, So it's not the case that I have a lot of factor returns for two months and then nothing for ten ten months, right, So that's not really a factor. And then possibly the third characteristic is that they have to be interesting, so they have to be in some way vaguely interpretable. So when
you you know you match these requirements, it's a factor. Now, now imagine that you have the Trump factor. Let's see if Trump wins, a few stocks will definitely benefit, a few stocks will definitely not benefit from the election of Trump versus Kamala Harris. Another source could be well tariffs, right, Another source could be AI. Okay, AI definitely right. Doesn't fit the characteristic of being pervasive because there is a relatively small universe that's affected by the AI theme is
likely not to go not going to be persistent. So it wasn't here like a few years ago, and it will probably not be here in five years because everything will be to some extent. AI it's interesting, but that's a theme, it's not a factor. That's what I would
call a theme. And there are also some mathematical characteristic of a factor versus a theme, which so basically you can create a portfolio that tracks a factor, and this portfolio will have a relatively small idiosyncratic risk, so it will be truly a reproduction of the systematic source of return that you were observing through the assets. So imagine that this systematic source exists, but you do not observe it directly. It's latent, it's out there, but you can
actually reconstruct it with a portfolio. A theme is let's say ten assets, you cannot really reconstruct it the same way because ten assets are just too few to diversify away the idiosyncratic source of returns of the individual assets.
So when you're like thinking about factor identification, how much of the money that you make the actual returns come from essentially factor identification or being able to identify a factor before other measure identify a factor that exists before other competitors out there in the market.
Okay, that's a great question because I I think I know the answer. Okay, great, But the reality is this, I think you know somebody else's factor is my alpha, and vice versa. Right, So say more, there are well known factors, let's say, some variety of value and momentum or reversion, and you can bet on those and you diversify away everything else, and what you get is, basically you get some returns that are priced priced in the sense that, as you know, you pay basically some risk
for that. Right, So this is priced return and that's great. But once upon a time like these were non not public knowledge. If you were lucky enough to be a hedge fund in the eighties, and I've met a few, you know, and you were maybe also investing in Europe, these factors were really working very well, and they were alpha. They were not called factors. You know. The first I
think published paper is probably eighty nine for momentum. Right now, there is alpha, and alpha is basically ideally would be a return that has no asocidate risk to it. It hardly ever exists. So what you really have are factors that exist at some frequency or in some universe, or with some characteristic that nobody else has found yet, and so they can be exploited.
More, how do you make sure that when you're isolating a particular factor, you're not accidentally taking into accounts some other dynamics. So, you know, maybe you want to invest in a bunch of companies with like pricing power during the tariffs, but actually your cohort of companies ends up just looking like a bunch of big tech companies or something like that.
The short answer without explanation, is that you can. But the long answer is a little bit more involved. If you have true characteristics, like I don't know, a tariff and a tech classification that are one hundred percent correlated,
well then you really have only one. You don't need both, right, So okay, But if I have in my let's say, arsenal of factors, if I have multiple factors they're somewhat overlapping but not completely overlapping, then you can build a portfolio that separates the impact of one from the other.
So you try to isolate you can you.
Can isolate them, you can kind of purify them. Now there is also the scenario where there are factors that are not in the model and they should be and and basically those they complicate the picture a little bit. But otherwise, if you have a reasonable model, you are you're going to be able to separate them to understand
what's the relationship. You can create a portfolio that exploits the first one, and then create a second portfolio that is uncorredly to the first one that exploits the second one.
Just zooming out for a second again. And this sort of relates to Tracy's first question, but also, I guess relates to my first question. If you have a fund and it has various pms and analysts in there, is there a difference between quant at your level, which is at the fund level, versus say a POD or a PM whose specialty is quant trading. And there are different definitions or different senses in which that term can apply.
Yeah, the fact is that you know, quant is is a very very generic label nowadays. Yeah, so there are many, many quants and they do all sorts of very interesting jobs. Some of them are are just differentiated because they live in different constructs. So nowadays, in a platform, especially in a quantitative one, it's not impossible to see pods and center groups. Okay, so that's one distinction. So what's the what's the difference. In a pod, you typically have a
siloed group. I'm probably not stating the obvious, but you know you have a silot group. They don't communicate with other pods. You want at the firm level to have independent sources of alphas, and their payout typically is a percentage of their P and L after costs. Okay, and then you're a quant in a pod. In a center group, typically you are part of a larger group and the group will hopefully have large capacity. So these have you know,
a larger program, like a larger research program. Their compensation tends to be more discretionary. And that's a center group. Then you have all sorts of other quants. So you have people like me who serve the firm at the center level. I also serve the leadership of the firm. And then you have people doing who are doing, for example,
execution research. She's extremely extremely complex and interesting, right, So it's not black and white like you can do execution research and be responsible for some P and L. It's very very very rich nowadays and very specialized.
I was actually going to ask about execution because when we're talking about quant investing, I think a lot of questions are around factors and idea generation. But you have all the I would assume boring stuff like liquidity trading costs that you also have to think about how do you actually incorporate those into your strategies.
So you can do it in a variety of ways. It depends first of all, what position the firm occupies in the ecosystem. So if you are a high frequency trading company, most likely you are using your own capital because you are capacity constraint, so you know you don't need a lot of capital. So those firms exploit market microstructure level information. Okay, so in a sense, a high frequency trading firm does not have a market impact model in the traditional sense. They don't see parent orders, right,
they execute at the microscopic level. If you are a hedge fund, typically you trade a lot, you have your own data set of orders. These data sets differ a lot, so you could have a market impact model for a quantitative trading group or a strategy, and you could have a different market impact model for hedging and a different
market impact model for fundamental investing. And then what you get is basically a term function that you place in your optimization problem that hopefully help to size the portfolio or trade the portfolio optimally. And this is extremely important. Uh you know, market impact is it is a very very sizeable uh fraction of the lost P and L of of a firm.
What as of today, what value is there in your world of specifically generative AI, l O, MS, et cetera. What how do you how do you currently or not currently get actual value out of them?
Okay, so on this I have really relatively little to say. That's that's original.
But tell us everything your employer is doing with AI.
Yes, that'll send you the resum, thank you, But I think, okay, just let's recap the basics. Right, So the basics are, at least for the time being, everybody is trying to be more productive with AI. Right, So you want to have all your documents you want to have now you know what? Perplexity has a finance module. I think one day soon maybe Bloomberg will not have the keywords any longer. You just give you know, Bloomberg a task and it will grab all the pieces of information and hand it
over to you and maybe you can schedule it. All of this is relatively table stakes. I mean, the the agentic aspect is not yet, but it will become pretty soon. I think it's going to be very hard to compute with the likes of maybe Bloomberg, but for sure, let's say you know the big hyperscalers. So that's one. At
the investment level, it's it's much more complicated. So in strategies where there is a natural richness in data, you can definitely use if not deep learning, but you or AI, but you can definitely use very advanced machine learning algorithms and you do not have a data snoopin problem, you do not have a back testing problem, and so you are in a data rich environment and you can do that.
And it's not a secret that, for example, XTX has a very large on premu you know number of in Nvidia cards I don't remember h one hundreds or something like that. So that's one thing, right. The question is really what's going to happen to the slower investment styles. And my view is that hopefully large firms like mine will have an advantage. But will see right why because
we do have we do have the scale. We have a large number of pms, We have a lot of historical data, we have a lot of propriety every data that nobody else has. So maybe that that will work out. But how to make it happen, I don't know because things are changing so fast. And also I'm, you know, relatively a tourist in the areas I'm trying to learn a little bit more about.
You mentioned proprietary data, and this comes up a lot where people talk about, well, the competitive advantage nowadays really is that data set? I mean, is is that true? If I get something really cool and unique, I can automatically become I don't know, a billionaire trader, if I can figure out how to execute on it. Is that all there is?
Maybe yes, I have very weak beliefs on this. I don't know. Maybe yes, we'll find out.
Well, so where are people getting interesting data sets from?
I mean, you get interesting data from observing human beings actually investing, and you don't get to see a great PM investing, but I do. That's that's the benefit.
So from your central position, you just get to see a lot of activity and you get to see novel data that other people don't get to see simply by being in the center of all of these different trades and everything, and that gives you a sort of higher abstraction layer or whatever it is that someone else in the market doesn't have.
Yeah, and it's possible that not in the distant future, good pms will become good because they can improve on themselves by basically playing or training or having a baseline of an agent that reproduces their behavior. So you know there is an alter gapy, well ano a PM, but an alter whatever who says what would you do right?
And you get a baseline behavior and then you can think about it and you could say, well, I would do something different, and then that becomes an example in a reinforcement learning process where the AI keeps learning from you and you keep improving because the baseline is changing.
So before we came out here, I asked perplexity to come up with a new factor, and it came up with something called the policy agility factor, which is supposed to be that countries that display policy flexibility have better outperformance over the longer term. Countries that are able to more quickly adapt to changing situations are outperformers over the long run. Can you grade that factor. I didn't do
a back tust. But like if someone brought you an idea like that, not me, perplexity, I don't want you to insult me over the next five minutes. What would you say to them? What are the problems with this?
I mean no major problems, there are questions. So the first thing that you want to make sure is that if AI whatever it means, brings to you definition, right, that definition should be at a point in time and should not be trained on all the on all the past data, right. So number one, you want to do that because if you back test that feature and in a way perplexity has already tested it, it's not a fair play. You know, the performance will the back test
will look great. So, unfortunately, we live in a world where some factors will never be back testable. So you don't know whether they work or they don't work, right, You just know that you cannot test them in advance, like a policy agility. This seems to be a very low turnover factor, right, and it seems to be probably a very low.
Sharp factor in a low universe and.
A small universe. So how do you how do you know? Well, probably you want to make sure that it makes sense, and maybe you can start putting a small volativity allocation to it and.
Then you would build it up as you.
Watched it out for yes, out of sample.
Okay, so speaking of back to us, I have one more question. But it seems like quant investing. Part of the issue with this is you are looking back at historical data. That's all you have. You don't have data about the future. Unfortunately, it strikes me as hard to
deal with regime changes. So when you have a big break in how something works in finance or markets or the global economy, how does quant investing actually take into account those sorts of risks, Like say, you know, a lot of investing is based on the idea that bonds and stocks are going to move inversely to each other, and then in twenty twenty two they started moving together.
I think that most people with a quantitative background in finance will tell you that regime change is very difficult to detect and to act on in an effective manner. So I think that's been my experience at least, right so in every possible application I've tried, and you know, it never, it really never works for me. Maybe it works for somebody else. What I think it's a bit easier to do is to detect regime change in a human being. So instead of trying to use you know,
there are many many algorithms for regime change. You know, there are MARKT based q sum, completely non parametric. Instead of trying to act on regime changes in the environment, try to detect changes in the behavior of a portfolio manager and act on that because that works, I think, and usually you know, jives with experience with so that that is something that can be exploited.
I want to go back to an answer you gave early on, which is sort of like the old school factor investing and like the original versions and maybe they sort of an international factor or a liquidity factor, or
the small cap factor, the value factor. And it feels like a lot of these things haven't worked in ages, and there's this debate that seems like, Okay, is this the long cycle and eventually it's gonna come back, or is it that everybody not only knows about these factors that have discussed them to death, they're also extremely commodified in the sense that you could just buy an ETF of them, right, You could just buy a small cap ETF. It's trivial to execute. You could just buy a momentum ETF.
It's trivial to execute a value ETF, et cetera. Like my intuition would be, since everyone knows about them and they're completely commodified technologically, they're just gone. But there is still debates. Some people think it's totally a matter of time before these come back in vogue, and that it's the long cycle, et cetera. I'm curious how you think about some of the original factors that people discussed in their prospects going forward.
Well, so some factors were identified, but then somehow they got demoted so famously. Size, right, so conditional on having other characteristics of a stock. Size doesn't really explain much of your returns, and so it's a combination of other factors. Okay, well that's one case. Then there are cases where it seems that some factors have been exploited. You know, their capacity has been exhausted, and so you can't make an
attractive return of them. There are some factors that still have a low sharp, but they still have a positive sharp, and so you know every positive sharp deserves, however small and allocation.
What's an example of.
That medium term momentum. Right, medium ton momentum is treadable and it's relatively high capacity. Then you have the whole term structure of momentum, so you know there is a shorter horizon reversal and whatnot. Short interest worked great until it didn't really work so consistently any longer. And then they also assume different characteristics, right, so you start having more crashes and the like.
Is there a mean factor?
I don't think so. But is there change any theme or something like that it's a theme or a theme? Yeah, I don't know that ESG is a factor either. I don't think so.
Oh why do you say that?
Because I don't think it's really that persistent.
I mean, it doesn't affect human behavior.
I think that just there is also this feature, right the moment that you say that a factor exists, it's reflexive, right, there is reflexibity in this, right, But I don't know that it really explains much of the returns in recent times.
So I'm going to ask one more question because I started with a dumb one, and so I will finish with another dumb one. Is there good and bad alpha or is bad alpha just beta? No?
Every alpha signal is you know, God's little child. There is no bad alpha.
All right, Gappy, thank you so much for coming back on odd Lots.
We're gonna leave it there. That was our conversation with Gappi Pallioligo. I'm Jill Wisenthal. You can follow me at the Stalwart, Follow Tracy at Tracy Alloway. Follow our guest Gappy, He's at Underscore Underscore Polioligo. Follow our producers Kerman Rodriguez at Kerman armand dash O Bennett at Dashbuck and kel Brooks at Keil Brooks. From our odd Lots content, go
to Bloomberg dot com slash odd Lots. We have a daily newsletter and all of our episodes, and if you enjoy the show, please leave us a positive review on your favorite podcast platform. Remember, Bloomberg subscribers get to listen to Odd Laws and free on Apple podcasts. Just go to the Apple podcast app and follow the instructions there. Thanks for listening.