How the Hottest Hedge Funds on Wall Street Really Manage Risk

more and more detail. And I feel confident that one day, after we've done like fifty episodes on this topic, we will get there.

had incredible runs. Really seems to be displacing a lot of the old style quant disrupting the sort of fund of funds idea that was popular. I have some sense, you know, you have all these managers and you give them very specific mandates and they have to really focus on that, and then if they're not too correlated with each other, you can get above market returns in theory and apparently in practice, but like how that actually works, I still really don't know.

And I have never gotten so many people like talking to me on the subway when they saw me pull out Advance Portfolio Management and they're like, what is.

things like Nvidia, we've seen small caps shoot up. A lot of people are talking about whether or not this is a factor rotation, and we'll get into what factors actually are. But I think the discussion that we're seeing right now, and I should caveat this with it is July eighteenth. So we've seen those big moves in the

market very recently. The discussion that's happening now is how much does the I guess growth in factor investing feed into some of these moves, and also how does the risk models that go alongside this actually impact investor behavior and then also feed into these market moves. So is it the case that everyone's getting out of big tech because their risk models are telling them.

for international oil plays. And then we know that like and then you're not allowed to take any sector beta, and you're not allowed to take any market beta and all these things, and so you see fact you're neutral factor neutral, and then you know tight risk limits. So if something starts to go down, you don't want to lose your job, and you like get out of positions, and that can create interesting moves for the market. Anyway, suffice to say, there is much more to learn.

see sketchiness in the market. So I think we should talk about, you know, what are the technicalities that are driving that pod shop behavior.

performing relative to their benchmark or expectations. So we're gonna maybe understand a bit more of the technical aspects of all this stuff. So Rich, thank you so much for coming on.

deployed at the pods. And then after that was at Viking Global, which is I always describe the Tiger Cubs in some ways as like a hybrid between the sort of equity long short value orientation sort of philosophically and the multi manager systems. And then finally, most recently, was a portfolio manager at Citadel managed a global materials, natural resources and materials portfolio.

a lot more and we can go into that. And then as you kind of migrate sort of down the time horizon spectrum, at least from what a thesis looks like on a single stock, what you're kind of doing is thinking about where are the catalysts that change the market's perception of that long term. So, like I remember when I joined Viking, I remember asking the question just generally,

like how much do you care about earnings. I think for anybody who's very value oriented, you kind of are concerned and about like, am I just going to be focused on the next data point, the next earnings and not sort of able to you know, see the forest for the trees and sort of care about, you know,

what does this data mean for the long term? But the answer I got back in general, not specifically there, but is in that kind of framework it's or the way the question was answered to me was, Hey, the long term is a function a DCF is a function of years. Years are a function of quarters, and so therefore we care about the quarters. But what that tells you is that, like the answer is what about the short term, catalyst changes the perspective about the long term

valuation of the company. And so I think what people sometimes looking from a far don't appreciate is the extent to which there's actually a little bit more of a convergence across styles from the underlying analyst workflow that like even a very long term investor to some extent is saying, even if I'm betting on the long term, the interim proof points illustrate the view of that long term and the short term guy says, well, I may get the number right in the short term, but that only is

meaning to the change in the markets price if it tells you something about that long term. And so there's a little bit of like a yeah, and I think that convergence is happening more and more where people are kind of pushing towards that center actually, where everybody both cares about the short term data point and is looking to what that means about the long term. So but anyway, at the beginning of that process, at the silver point or any deep value type place, you're just really focused

on that longer term story. You're less focused on the quarter or the catalyst than trying to understand sometimes things that And I was a junior analyst when I kind of started there. There was a first job out of banking, and you know, but you can be looking at like what does the rail contract look like in twenty twenty four and how does that step up? And you're like, man, does this matter to the stock. It's great training. It's a perfect place to kind of get that.

about fifty to fifty split. I would say between people who I call like natives who come from the risk model system either any of the major pods or related and the other half can be like a deep value fund that says, hey, I don't want to limit myself to this stuff, but I see just like you guys are saying, this is an increasingly important part of markets and I want to be deep on it, educated whatever.

So to answer your question on how multimanagers kind of pickstocks, run processes and think, not specific to any one place, but that sort of natives group in general. So at any of these places, the core contract is to say, the core difference, I guess is the other way to say it is versus a deep value place, it's about turnover. In the end, it's two things. It's risk limits and

it's about how freak your book turns over. So at a deep value fund, the goal might be in sort of theory to have more than a year long average hold period. In practice it'll be often shorter than that, you know, nine months or whatever as bad idea cycloud or whatever. But at a multi manager those numbers can be anywhere from like ten to fifteen to even higher, meaning the entire book turns over ten to fifteen times in a year.

a sort of defined single universe. And then a portfolio manager will have a set of analysts below them who have typically very related coverage universes and will feed up into the portfolio manager. So that's like kind of the structure. Stockpicking kind of ends up being what we were talking

about earlier. In the end, it's the analyst job to have a detailed model, of course, to have a view on earnings across their coverage universe, and that coverage universe for that analyst, by the way, can be and it varies by a multi manager, but it can be anywhere from like thirty names at the low end to like eighty names at the high end buy analyst. So there's a lot of process there.

it seems like a much more sophisticated process. So I guess I'm curious if you're at a podshop, how do you know how much to buy? How do you know how much to allocate to a single stock. And I guess another way of saying it is you're looking at that single stock on a risk adjusted basis, right, Like, that's what you want to get, right, the risk adjusted performance, not just the single stock performance.

The third level is factor neutrality. I'm balanced against all of these sort of if you maybe simplified just slightly the subcomponents of beta, so instead of like, hey, I have a beta to the market, I actually have a beta to the basket of size large companies. I have a beta to the basket of companies with momentum. I have a basket to the beta.

non factor component of any stock. So as you look at any stock, it fits within that broader portfolio that you're putting together.

which would add to your execution costs. So does that come into play as well?

trading costs? The engines operating at the multi managers do think a ton about how do I take the stock picks that a single portfolio does and then execute them in a optimal way a crossing some firms doing some prints don't cross each other's orders within the pod level about all of that, and then so the first level of how do people get limited is the constraints on what percent of my bets are in factor type bets

versus non factor type bets. There are all also a bunch of like single position limits, So that's like one version is basically limiting the portfolio manager to have to

sort of live pickstocks under this constraint. The other framework of how do you size positions to your sort of earlier question which comes around to this trading cost question is there are tools that are called optimizers that basically look at the expected return that each portfolio manager thinks they have in their book of stocks and tries to solve for the optimal balance of the expected return against the volatility of those stocks and the volatility the factor

bets in the book, and it'll spit out an answer for you. That answer may not be exactly where you want to land, but in the most sophisticated places, that answer that the optimizer spitting out is including how much trading cost impacts the book. So it's sort of flowing that mathematically into a machine driven optimal book. But again that's sort of in the more science bucket. Of course there's art even underneath that statistics, but basically that's in

the more sort of science bucket. Then the PORTFOLI manaddressed to say, Okay, the machine sort of took my expected returns, took the variance of those pieces and the trading costs into account. I gave me an answer. Does that actually still fit with my you know, fundamental bottoms at work? Back to hey, the contract of this company changes in twenty twy six, the earnings going to be this. Here's the positioning and set up and crowding of other pods,

you know, playing the game you mentioned earlier. So there's then the sort of second level of art that goes to the top of that.

specific idiosyncredit. That's exactly right, right, But why does that inherently lend itself when you're thinking about I mean, I feel like there must be some connection, But you're trying to strip out all of these different factors that you don't want to have exposure to. You're trying to find the idiosyncratic drivers of a specific name. What is it about that process that sort of inherently lends itself to shorthold periods.

is approximately normally distributed across a certain window. Meaning it's sort of, you know, like flipping a coin basically, And the intuition is, if you flip a thousand coins, obviously you'll center around whatever your hit rate is on that coin.

If the coin is loaded fifty two percent, yeah, versus fifty As you flip three coins, it could be you know, the mean, the expected value of that is going to be you know, who knows, right, But as you flip a thousand coins or ten thousand coins, you will center around that mean. And so and that variance is effectively, if you think about things from a return standpoint, the sharp ratio, right is they're returned about it about the

variance of the volatility of that return. And so as you have more and more bets, you shrink the variants relative to the return you're generating. And the more and more your bets are in idiosyncratic bets which are normally distributed unlike market bets you know, which are you know can be wild.

period of time. Okay, so that's kind of also what the model solves for, and it sort of solves for a combination effectively of what's the highest R squared meaning how much of the model explains what's happening across stock movements across different stocks in the market, and then the output of any regression within its period is going to produce that result of a normally distributed kind of residual term.

But the key way that this model works is that it's normally distributed, not across time but across stocks within a given period. And so what that means is you're going to have you know, as many stocks that are on the residual basis that are outperforming in a period

as that are underperforming on this residual basis. Whereas, of course, if you just bet on semis right in a month, and you just were long Semis within a period within a month, right, that's not going to be normally distributed, of course, Right, It's just if you're managing to a model that is cross sectionally approximately normally distributed within a month, Let's say you're going to get winners and losers, and you're going to center around that hit rate basically.

month horizon or a six month horizon or whatever. And so you're trying to pick the calibration horizon that matches the investor that we're talking about. So I mentioned a month because a lot of the multi managers, let's say the average hole period ends up around a month, you know, twelve that's twelve times turns a year, but it could be high. It could be seventeen turns, it could be eight turns. They are managers who are in that range.

Totally that led to that idea? Well, not just that they like the stock, but that they like the stock in a very short period of time.

slower too. But anyway, with with those caveats to your question, yeah, so there's in an ideal world you do a you sort of separate the idea generation process into two steps.

I mean, you understand, you look at industry earning earnings and earned industry reports and filings and all of those kinds of things. You talk to experts also as part of that process. That could be any of the expert network calls somebody who were people who were executives, and that can inform that initiation and understanding of the industry

as well. And some people spend you know, some analysts spend the majority of their time doing sort of initiation type work that sort of build a deep financial model that tries to build not just from like the high level revenue but to unit economics like okay, And by that I mean like, you know, if you're looking at a coffee shop, like, okay, how many cups of coffee do this? Hell, what's the price? How much is that going to change? What are the inputs to a cup

of coffee? And just trying to get to that level of granularity on unit economics. Yes, and so that's kind of like the initiation process, and then ongoing coverage is a little bit more of Hey, I have a view from that initiation work on sort of long term relative winners and losers in a space. I have an understanding of uniit economics of each player and how each of

those is kind of heading. And then the ongoing maintenance process is a lot to do with what data sets, what data points, what conversations from an industry conference standpoint or whatever can I do to understand more granularly how

each of those unit economics points is changing. And then finally also like there's this question of crowding and positioning and understanding what everybody else thinks, that sort of weighing machine versus voting machine ben Gram classic analogy, but you sort of separate that process, have that secular view, and then you're trying to understand what data sets. So that could be like all data sets, It could be industry conferences, it could be talking to people in the industry through

the supply chain. It could be you know, people always should be doing this but don't always actually in practice doing it. But your analyst should understand if they're covering an auto company, they should understand auto suppliers, and they should understand the downstream of that. So each of those

sort of up and down the value chain. That's like a big I'd say in reality, a differentiator among analysts is how deep into the value chains you're seeing what's happening to inform you about the changing trends in those unit economics that you had a baseline view about at the beginning of the sort of initiation you understanding the industry.

you can be long. Ever, so you're limited in your ability to be long, and momentum can have nuance like do you calculate momentum over a six month window a nine month window and what are the inputs to that, But in aggregate, you're actually limited in your ability to be long or short momentum at all. It's actually one of the most focused on factors within commercial factor models that everybody asks about all the time. So that's like

point one to mention on momentum. The other is the way you described at the beginning was interesting too, because there's a concept of factor investing where you're betting on the factor, meaning you're finding cheap ways to be long momentum or cheap ways to be long the value factor or other pieces. And that's the kind of growing and that ties into the whole sort of growth of passive

and all those things. What these risk models actually do in the multi managers essentially are the elimination of factor bet meaning it's the opposite. It's kind of a mirror image of that, where you're sort of eliminating the factor bets entirely and trying to find just the performance in

the residual. That then leads to this question of like what factors exist inside the residual term that are not momentum And that's where you get the concept that you mentioned earlier, like a pods blowing up and what's positioning and crowding and nuances there, which something we spent a lot of time thinking about, Okay, like how do we mathematize how do we characterize that? And what gives information incrementally beyond? Okay, you've eliminated this sort of straightforward momentum topics.

You've eliminated did value What within that residual can give you more and more insight beyond just like the core research work can we talked about earlier.

Here this factor is less well constructed by this model, and kind of put them together. And then there's sort of also a spectrum in terms of people software tooling that they how far down they hand into the organization a sort of sophisticated tool to let portfolio managers see what are my factor exposures. So like some places there's a total separation almost of church and state of you know, stock picking and risk management, and that is partly a function.

There could be a philosophy element to that, and there could also just be a constraint. I mean, it takes engineers and time and money and focus to build all this stuff. So some places will have nothing in terms of tooling, and they'll just have a risk team that kind of looks at books and helps people understand their risks on a sort of shorter cycle meaning longer cycle, like it'll take oh, once a week, once a month, whatever, they'll get a report on their risks, or they'll check

in et cetera, et cetera. And then at the far end you have funds that have like full software platforms that hand and to it portfolio manager like okay, if you change this, what happens to that? If you want to sort of see what the optimization math does for you instantly, can you see that? And so that's kind

of the spectrum of what things do. And we sort of provide that software toolkit everything from the risk model as you mentioned, like the core underlying factors all the way up to the software infrastructure that lets you just play with it. Okay, if I had a billion dollars in video, what does this do to my risk numbers, that idio number of fact number? What is it do to each of my factor exposures? And then how does

that change dynamically? And it'll also sort of like find hedges for you, Like what single stocks would optimally hedge this book in this way. Now, of course it's still on you to pick stocks, but it it'll source. Okay, I've got a whole universe of stocks. What single stocks would offset this in video? Or these five single stocks would offset that?

something for a car or whatever. But then what do they see to say and now we should buy it, Like what would be the signal that they're looking for in the market that say, you know, again on some short term period. This is really I've gotten to really know this company, but there's something about X right now that makes it a compelling buy for a short term period.

the market it's, you know, whatever the market means. But typically the market is who is the marginal price cetter of a stock basically, and there's a perception there implicit in the price at a minimum about okay, how many units of coffee and what's the price of those coffee cup's going to be, and what's the underlying cost.

positioning comes into the equation. And everybody's position this way, and I think it's going to go the other way, and I've got a very tactical thing that is a part of the equation, but a much larger part of the equation are still catalyst driven, Like Okay, there's a data point that comes out, but it's a data point that indicates something about the overall perception of where this

company is headed. And so like classic ones and software can be changes in churn direction and like where people can get smart on that is often like okay, there's an overall headline churn number, but then there's like like if it's an internet company or something like that, or subscriber company, and then you can go down the line like okay, if somebody's looking at churned by region and has some forward look on something that gives them insight

to like, okay, churin is changing in this region, and this reading is small today, so it actually doesn't hit the headline churn number. Yeah, but that's actually structurally growing faster than every other region, and so the underlying churn rate that looks like it's this level is going to step up structurally because this smaller region is going to be a bigger part of the overall path. That's the kind of thing.

there's you know, it's important. You mentioned the different shops have different software infrastructure, and the level at which it's on the managers different sometimes and the different which it's at the umbrella level, So like does that mean that, like does it happen where the at the very high end of risk management they look across and they say, wow, you know, in aggregate, our portfolio managers, maybe perhaps unintentionally or even within their remit, have built up a lot

of implied exposure to momentum or implied exposure to rates, or implied exposure to value. And then what do they do, like tap people on the shoulder and say, like, how what happens?

is monitoring each level of that. So like you'll put limits, as we talked about on the portfolio level, right on an aggregate risk basis, and then on an individual factor you'll say, okay, you can have more than blank percent of your variants in your book in a specific in any specific factor, So put those limits individually, and then exactly as you said, they roll it up just like

you know, you just add up the line items. Essentially, all these models are structurally linear decomposit so they add up actually linearly, So like John's momentum exposure in dollar terms, here Jaill's exposure is there, and they add up. So you do see aggregate level CIO level kind of hey, we're net lung blank or whatever at that level, and it depends how teams structure their limits and how tightly

they limit exposures at the portfolio level. But you will see aggriate exposures, and then there are ways to take like an ETF or a basket or a custom basket that will just limit out We'll just literally hedge that basket. And there's nuance even there, like, hey do I can I build a basket that hedges out that exposure but doesn't actually basically end up being short at the same stocks I'm long underlying the book, you can see how that can get into a whole rabbit hole of like

sort of technical behind the scenes execution detail. But at the high level, you sort of roll up the exposures, you add them up, and you say, am I long or short one or two or three or all the factors, and let me balance those out at an aggurate level.

and the model spits out and says, everyone needs to sell right now. Is that a risk? Is that like a realistic risk or is it the case that all of this off the shelf risk management software is so customizable, I guess, and there's still that discretionary factor for the PMS that you don't really get that hurting behavior.

they're kind of proxies for each other. And so kind of what the model lets you do is kind of, instead of having to be perfect pairs in the Alfred Sloan original hedge fund concept, where you have to in order to be factor neutral, you just have to find perfect comps, it kind of lets you pick non perfect comps but end up in a risk place that is similar to that where your only bet is on a

single stock. But anyway, so like what the model is pushing you to is not any specific stock, right, It's telling you to pick which one of the stocks that don't have comparable factor exposures is more attractive. So that's one level. The second is there is leverage, and the

leverage you're putting on is not leverage against beta. That's the distinction that I think people often allide is that when you think of like LTCM or maybe forgetting even LTCM, but any fund that takes very high leverage on a beta a directional bet that's beta on a factor, and the issue with that is many issues with that. If you're taking lots of leverage on a beta is there's just sort of that risk that it has a big

draw down. The hope, I guess, or the sort of mathematical reality as you kind of pointed out that's actually been executed on is that when you're levering alpha, it's again it's sort of the quant fund world works this way too, is that what you're levering is just that residual term you're getting back to that coin flipping and you're finding a source of return that is normally distributed across stocks, and therefore if there is a big blow

up in markets, actually typically the factors become more and more statistically significant, and so if you're neutral against those factors, the residual return remains a cross sexually normally distributed. So there's obviously a lot of detail under the hood, but the basic answer is that you're trying to find a type of return, and a diversified source type of return that doesn't have that risk in a blow up, So you're kind of levering alpha. That's the key kind of

point versus beta. And the final yes answer to your question is that you are still levered. So notwithstanding everything you can do to sort of solve the mathematic piece of this equation, you still have some risk that the person providing you the leverage has a business problem or somebody who like whoever is providing that leverage to you,

which is typically the banks. Basically that person for whatever reason needs to pull that leverage or whatever, and that it's almost a little bit even in the category of business risk that exists intrinsically with leverage. So that's kind of the the yes portion of the answer.

be at least have something. It may not be the only model that's viable to make attractive returns for investors, but at least that sort of you know, result has jumped from the sort of academic theory to realized practice.

that jump long ago. It's in the fundamental stock picking world that it made that and again, a few firms had been doing it for a long time, but it sort of made the most convincing leap over the last whatever five to ten years, where it just sort of decisively generated very attractive risk adjusted returns for investors and kind of proved that sort of synthesis, which is really what's happening between the sort of quant view of the

world of factorization and finding idiosyncratic or residual performance within inside what's left over after the factors synthesized that risk and sort of quant perspective with that Warren Buffett's style fundamental type research and analysis and work, that synthesis was implemented by a few firms and now it's sort of proven itself to work in a lot of different ways.

So that's what's happening. So from our angle, like what we have seen is just that wide range as I kind of mentioned earlier of execution of that, like how easy is it to actually have a system in place for a portfolio manager or analyst or cio. How sort of not only user friendly but sort of functional. How efficiently can it source new hedge ideas that balance out a specific factor exposure? How efficiently does it connect that

risk perspective of where I'm long and short? To a topic you mentioned earlier, performance attribution, like where are my generating returns? What are my hit rates, What are my hit rates on residual versus on factor? What are my hit rates on earning season versus outside of earning season and on a residual basis, And how does that connect to my risk and my portfolio construction? And all of that is a lot of work, you know, it's a lot of painful kind of putting together the software and

the risk and all the different elements together. And as I mentioned, what we see is some funds have done this, you know, at a level that is really excellent, and some funds, most funds, because they have to do the very hard work of stock picking. It's a very challenging job. You have to have incredible IQ allocated to that problem

and effort. They have, Okay, systems of a sheet that gives some risk insight, but it doesn't have the sort of detailed input output experience of let me tweak this, let me see what happens there, let me understand how it connects. And so putting all those elements together is what we kind of hope to do.

of like a steady escalation in thoughtfulness. I would say, as you hand the portfolio manager tools on this factor, and you also sort of empower them because again and a lot of these organizations are set up where there's a risk side and a portfolio management function, and you know, the portfolio manager isn't necessarily the client of the risk

in house at these places. But as there becomes this industry of people like us who are providing these tools, in a way we have to be a little more responsive to the portfolio manager says Okay, I see how that was built. Can I double click?

a constraint in a lot of cases. And the best places are doing it where it's completely synergistic, where you're using the risk tools and this factor awareness and all of the things you can do with that on offense, not just defense. So that is happening at a few places, but there's a whole other group of places where it's kind of, hey, this limits me, This isn't working with

or for me. And so as this becomes a little bit more, you know of an industry or like, we do work for them, right, so we're hey, I want this additional feature that you know they're the client. Right.

So there's that piece, but there's sort of this constantly escalating sort of demand for tooling and incremental insight and okay, let me click this, let me understand this across my whole universe, across the entire universe stocks, I could cover all those kinds of tools on the sort of the people who don't come necessarily out of the POD systems. The interesting thing is the extent to which people want

to focus on. Okay, let me, how do I frame that system that factor awareness instead of in a market neutral context. But hey, I'm a long only or I'm a sort of directionally oriented value fund or whatever. How do I reorient the model shift it to be sort of true comparative to my benchmark? And so that's been an interesting evolution is the types of investors who are not structurally market neutral but still want all the insights from this where you can recalibrate the entire model against

a benchmark. And so that's been one example.

And then Rich was talking about how when you're coming up with investment ideas, you're sort of trying to identify something that will change everyone's perception of the trajectory of a particular stock or investment totally.

causes the you know, the whole market to crash. I guess big events don't usually happen to cause the whole market to surge, Unfortunately, it's always the other way around, Like nobody knows what interest rates are going to do, and we know you know, a lot of stocks are tied to interest rates, and no one knows maybe some chip company will come out tomorrow that beats and video whatever.

No one knows any of that stuff. And then this idea that if you can then strip out all of this and then identify the idiosyncratic drivers of a stock, and then those idiosyncratic drivers of stock almost inherently, some will be winners and some will be losers. Yeah, I could see why. Then the game is lots of bets over relatively short time periods. Like that that like really clicked to me in this conversation.

your source of alpha. But again, the thing that's coming through in this conversation is really like the diversification aspect, the desire to be factor neutral and to lever the alpha instead of the beta.

to understand more about that. I don't know, there's a lot more that we can also, just like the different models, Like I'm sort of fascinated that, like there's all of these different multistrad funds that exist, and the fact that they're not all the same is interesting to me. And the fact that like where the risk manager sits in the amount of tools and what they build in house and what they don't, and the degree of flexibility that pods get and you know, what analysts actually do and

stuff like that. There's much more to do.

a blog and a newsletter. And if you want to talk about all of these topics, including investing and markets, you can do that twenty four to seven in the odd Lots Discord Discord dot gg slash odd lots.