Andrew Beer & Adam Butler - Attack of the Managed Futures Clones

Hey everyone, Corey here. Thanks for tuning into another episode of Flirting with Models. If you're enjoying the show, I'd greatly appreciate it if you'd take a moment to rate, review, and most importantly, share with a friend. A word of mouth is how this podcast grows. And if you'd like to learn more about newfound's platform of return-stacked mutual funds, ETFs and model portfolios, head over to returnstacks.com. Now on with the show. Are you ready? Yeah. Okay. Three, two, one. Let's jam. Hello and welcome everyone. I'm Corey Hofstein and this is Flirting with Models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy. Corey Hofstein is the co-founder and chief investment officer of newfound research. Due to industry regulations, you will not discuss any of newfound research's funds on this podcast. All opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of newfound research. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of newfound research may maintain positions and securities discussed in this podcast. For more information, visit thinknewfound.com. In this special episode of Flirting with Models, I'm joined by two guests, Andrew Beer of DBI and Adam Butler of Resolve Asset Management. Rather than my usual interview format, I wanted to foster a conversation about the replication of managed future strategies. Specifically, I wanted to bring on two practitioners who both share the same high level beliefs. Namely, that more investors should allocate to managed futures, that managed futures are well suited for replication and that replication can help dramatically reduce fees, but who differ on the implementation details. And it's in that disagreement that I hope to highlight the different pros and cons, as well as any embedded assumptions in any of these replication approaches. We discuss return-based replication, process-based replication, determining the number of markets to trade, expectations for tracking error, and much more. I hope you enjoyed this episode with Andrew Beer and Adam Butler. Gentlemen, welcome to the episode. This is one of a kind for Flirting with Models. I don't think I've ever hosted, I guess I'll call it a friendly debate, friendly discussion, meeting of minds. For those of you who are tuning in for the first time, hearing Andrew and Adam, they are both past guests. I highly recommend you hit pause. If you go listen to their previous episodes, it will provide a tremendous amount of illumination to, I'm sure a lot of the stuff they're going to say today, but this is a discussion that was born out of some Twitter conversations. Andrew, you recently published a piece in institutional investor all about the factor replication of the SOC-GENC TIA index and managed futures at large. And I have to get some disclosures out of the way before we go a little bit deeper. I am co-sponsor as well as co-sub-advisor with Adam on two ETFs that do managed futures replication. And my firm newfound research is also a model provider who allocates to Andrew's ETF, which does managed futures replication. So I am all in fully biased here towards both these parties. I will do my best as host to remain as independent as possible. But I wanted to bring you both on with the idea that we all collectively believe managed futures should be a bigger part of everyone's investment portfolio. We all collectively believe that managed futures can be replicated as a category and there are potential benefits in doing so. But we have differences in the way in which we do it. And I thought that would make for a really interesting conversation. So let me start maybe with a softball here, Andrew, just to sort of get things going. And I know you will knock this one out of the park. But why bother replicating at all? Why would an investor actually want to replicate managed futures index exposure versus buying an actual manager? Let's start with what we mean by an index for managed futures. So there is no way to pull the trigger and say, I just want exposure to managed futures the way you could buy an ETF directly on gold or you could buy futures contracts on gold. So when we see managed futures, what we're really saying is how have the guys who do this for living done over time? And can we use that as a guide as to how we think they'll do going forward? And just like nobody looks at the S&P 500 and says we're just going to ignore all the history of the S&P 500 to try to figure out how we think it's going to do over the next 10 years. So the return characteristics of this space to me is the single most compelling alternative you can put alongside a portfolio of stocks and bonds. If you look at this, the data starts to get good in the 2000s. All hedge fund data gets pretty murky before that even the early 2000s are a little bit iffy in some ways. But basically here's a space where you can look at it and say it's got zero correlation to stocks and bonds. As a group, it's not that risky. It's at a max drawout on a 15% over the past 23 years. It tends to do the best when you need it the most. So an index of hedge funds was up 20% last year, it was up during the GFC, it was up during the dot com crisis. And it has returns over time somewhere between stocks and bonds. If you're thinking from an asset allocation perspective, we just haven't found something that's a more valuable third leg of the asset allocation stool. It's just been kind of tough to get access to. And I think what everybody in this call is trying to figure out is how do we help bring this really, really valuable tool to advisors all over the country so they can help their clients make more money over time and sleep better at night. It's a win, win, win, win, win, win that we can figure out how to do this. So Adam, maybe I'll toss you sort of the same question, but more specifically to the replication side, why replicate in index? Why not, for example, just build a generic CTA? That's one that actually I mentioned that I was having this conversation on Twitter and I asked people for some questions. And one that comes up all the time is why bother replicating it? Like, why not just build a generic CTA? Well, the answer is because there's such a diverse number of ways that you can implement a managed future strategy. I mean, traditionally managed futures was relatively short-term breakout strategies, right? So you're kind of buying a market if it goes above the past 20-day high, you're shorting it, it goes below the past 20-day low. Over time, the breakout strategy lengths kind of expanded the turnover ended up going down a little bit. Or time they introduced a variety of other ways of measuring trend. Some of them simpler, some of them vastly more complicated. The industry is constantly innovating. There's constantly adding new markets to trade and new ways to trade them. And I don't think we should discount the creativity and innovation in that space by saying that I know all of the conceivable methods that they might use. And therefore, I can just bring those to bear in some sort of static or classic trend following strategy, right? So the idea here is instead of that, let's think about the different components of a trend strategy, the general big muscle movement markets, most liquid ones that they might normally be expected to trade. And then let's see how close we can get to the actual day-to-day week-to-week and month-to-month performance of the index taking advantage of all of the different innovations that they have made along the way. And using certain types of replication methods, the innovations that they will inevitably make in the future. So we've mentioned that both of you apply replication approaches. I think the meat of this conversation is going to be not just where you agree but where you actually disagree. The difference is in how you went about hitting the replication in your implementation. So maybe just for some table setting, Andrew, can you start with at a high level explaining your approach to replication and why you chose that approach? Sure. So let's give you a little bit of history. So we were sitting around 2015, 2015, trying to figure out and we were looking at the space. We could see the diversification benefits of the space. But it's, I guess, continuation of answering my first question is it's hard to get access to it. And pick one guy who specializes in more particular kind of strategy. He'll do well one year, not well the next year. If you go to hedge funds, it could be expensive. So they're all sorts of like hurdles to investing in the space. And when I was working with my partner, Betias, and a third guy named Matt Grayson, who is, you know, he's the math guy who solves 70 year old math problems, who just kind of operates at a completely different level. And his philosophy was try to do things as simple as you can and start, right? And do them in the most simple and most naive way. So we started looking at this and wrote this paper recently about replicating it with four factors. I mean, we started with one. Can you do with one factor? Can you do it with two? Can you do it with three? Can you do it with four? And it was just these sort of building blocks of starting with as simple as it could possibly be and then figuring out how we wanted to expand from there. So what we do is basically in an ETF that we manage is we look at the recent performance of an index of these guys that is published daily. And what we're basically saying is, I mean, if you think metaphorically about what these guys do, they've got these things, I consider them wave detectors. And they're pointing these wave detectors out over the ocean. And if gold is going up or if gold is going up, they're going to be a lot of things. down, they're trying to make an educated guess as whether it's going to keep going up or going down. And if it's going up, they're going to buy a future contract. It's going down. They're going to short the future contract. But my experience in what we've seen statistically is that waves move in clusters. You don't have things-- I mean, if you look across a portfolio of 50 or 70 or 100 of these things, they're doing one of three things. They're either going up. They're not doing anything or they're moving down. And when they go up, they tend to move up. A whole bunch of them tend to move up together. And when they go down, so in a sense, what replication does is says, we're not going to build our own models. We're not going to try to compete with these guys in terms of what they're doing. We're just going to shine a risk model at what they're doing with a very, very simple portfolio. We have 10 core things that we look at. And basically, say, today, what's our best guess as to how much exposure they have to these-- well, I think what you guys would call the muscle movement or the big muscle movements or something. So we just do the same thing. And what we found was that it actually works surprisingly well. I think a lot of people in the space think it shouldn't work well. But it really-- honestly, probably did better than we expected it to do. So we've really continued with that philosophy since then. Adam, same question to you. High level, again, just to set the table here. Can you discuss your approach to trying to replicate these indices? There's two broad approaches. When we call top down and when we call bottom up, the top down is very aligned with the way that Andrew describes their process. I think about it, if we were playing a video game, the top down process is I'm going to follow along behind an expert player. And when he shoots, I shoot. When he moves, I move. And there's sort of like a lag between what you observe, the expert doing, and then what you try to copy. The bottom up approach is, well, I want to observe the gameplay and the tactics, the general behavior of that expert player. And I want to learn from that. And I want to apply similar techniques. And I kind of want to then go off and do it on my own. And what we found is that there's a mix of both the bottom up, learn the gameplay, and the top down, follow along behind and do what the expert does. The combination works better than either one on its own. And we think there are good theoretical reasons why there are nice combination, because there are pros and cons of each method. And one method complements the cons of the other with their pros. And we observe this empirically when we sort of put the two together, they are not perfectly correlated. In general, historically, the bottom up approach, having an understanding of how the big trend following funds typically trade futures, the signals that they use to be long and short, and to manage risk and exposures in the portfolio. Replicating those basic mechanics does a little better historically over the long term in terms of tracking the CTA index. But there are stretches, sometimes lasting many months or quarters where the top down approach is more effective than delivers both better performance and better tracking of the underlying index. So the hybrid approach just seems to make great sense. So I want to start with this top down. And the other way I like to think about this, I often call them top down bottom up. But the other way I think about this is returns-based replication versus process-based replication. This sort of top down is let's look at the returns and be agnostic to how those returns are being generated and just try to find the portfolio that replicates it. Whereas that bottom up is, let's try to figure out the process by which they're actually generating these returns and implement that process. And I think all of us on this call agree with that top down approach. So that's common ground. So let's start there. And Andrew, I want to start with this article that you wrote an institutional investor that demonstrated that you could use this four factor model. Having one futures market from each of the major categories of equities, rates, commodities and currencies and we're able to replicate to a pretty significant degree the historical return profile of the SOC Gen CTA index. Again, with a very simple model. Now in practice, you use about I think a dozen, tenor a dozen. It's ten factors. You end up with about 14 positions overall. Adam, I know you use a little over two dozen. So I think it's 27 futures markets. And you're going to start with you. How do you think about this idea of trying to identify the right number of markets to use and which markets to use? And some of the trade-offs in that decision of increasing or decreasing that number. Sure. I think philosophically we are very weird for people who run quant funds in that we don't think about things. And part going back to this guy who, and his philosophical underpinnings, is, so our goal back then was to build a model with maximum robustness and durability. And what that meant was that if we set it, the measure of success is 10 years from now, the same model is working. The more you have to tinker with it between then over the course of the 10 years, you only tinker with things when they're not working. People don't say, oh, it's working as well as you wanted it to work. And now let's change everything. So in a sense, tinkering to us was-- so again, we started with really, really, really simple stuff. Just tried to understand it. Now, again, what's also a little bit non-quanty about what we do is I'm the third wheel on this team of quants in that I'm not a quant. But what I do is I do hedge fund style due diligence. I called every quant that I knew, guys who had been running commodity corporate research in the early 1990s. And what were the observations that all these guys had? I talked to guys who ran portfolios of these things. And so I think the general observation, we didn't find anything that contradicted what we were looking at. And so we started with these incredibly simple models, similar to what we published in institutional investor. But we were very, very wary of the worst failing of quants, which is to not see your own biases. And so we knew that if we tried to pick the perfect four-factor model back then, that's not robust. It's not going to work at some point. And you're making a much bigger call than you realize at that point. So one of the tests we ran was we realized early on that if you're going to do this well, you do need exposure to the four-major markets. So you need exposure to equities, rates, currencies, and commodities. And anytime you take out one of those, having eight commodities doesn't work nearly as well as having four instruments spread across the four-defernacic classes. So when you start with that, then we said, all right, let's see how bad we can be from a factor selection and still do OK. So we ran-- we took-- I forget how many contracts for each major market. And we ran a zillion combinations of eight different factors. And every single one of them did OK. Like, the worst was still modestly better than the SOC Gen-CTA index net of fees. And that, for us, is-- that's a measure of stability and robustness. That made us think that no one thinks it's going to work perfectly over time. And they're going to go through periods where it does better worse for lots of obvious reasons. But that suggests that the fine tuning that can go on and break is not likely to happen. And that bet, when you look from 2015 to '64, when we did it, it's been right. So in that context, we've never felt the need to expand beyond what we do. In fact, the hard part is not expanding. Because three or four times a year, there's always a moment where you're like, I wish I had something else. We wish we had something. If what if we had done this, the hardest part is not doing it. You and I were having that conversation earlier this year talking about, I think, was Mexican peso, JGBs, and egg contracts, right? It's so hard not to change. It's so-- you desperately want to do things too. But then you have to ask yourself, is the world really different today than it was six months ago? This was working six months ago, and a year ago, and two years ago, and three years, and four years ago. And at that point, your human beings, you have to make a judgment call. Do you change things or not? And our bias, which again, is very unquantied, is we're only going to change it when we've probably been shown that we should have changed it much earlier. But all those times that we didn't change it has probably worked to our benefit. So Adam, you were able to start with sort of a blank sheet of paper when designing your model. You ended up again just south of 30 contracts. Again, focus on the major liquid contracts. I think one of the things we're not discussing here, but perhaps we can get to later in the conversation, is these are packaged in an ETF wrapper, and there are inherent potential limitations. When you talk about the ETF ecosystem, about what future contracts you can and cannot include, 30 is a very different number than 10. And I'm curious how you got to the number of contracts, and what your thinking was about what markets you didn't didn't want to include. Andrews, right on, you need to include exposures to all of the major sectors that CTAs trade. So you need to include currencies and equity indices and bond indices and obviously commodities. And as you look across, I mean, we've got hedge fund products that trade over 80 markets. But many of them are not liquid enough to trade within an ETF wrapper and to trade at scale. So the first round was, let's look at all 85 markets or so that we've got good data for that go back a long way. Let's first of all take out the ones that sort of only started trading in the mid-nots. They don't have a long enough history for us to be useful. Let's remove the ones that are not maximally liquid. And in the end we settled on the sort of 27 contracts that they had history back to 2000, so the start of the CTA index, plenty of liquidity and they spanned the full universe of major bets that CTAs will mostly take. And then there are constraints, there are statistical constraints on especially the top-down methods. So revisiting the mechanics here of the top-down approach, you've got a basket of markets in Andrew's case 10, so let's use 10. And they represent currencies, equities, bond indices and commodities. And you're looking at the daily returns of the CTA index going back from, call it three weeks to 12 weeks. And you're trying to find a portfolio, the long short portfolio of our 10 futures markets, that when you hold that portfolio stable over the past three to 12 weeks, the returns of that portfolio look as close as possible to the returns of the underlying index. The way to do that is, well the simplest way to do that is via regression, so you're simply finding up the portfolio that minimizes the square daily differences between the tracking portfolio returns and the index returns. And to do that, there's some things going on in the background. One of them is you've got to find the covariance between each of the different markets, futures markets that you're using to track the CTA index. And if you've got only 20 to 40 different observations, and there was only using 20 to 40 days of returns in order to build that covariance matrix, if you have more than 20 or 40 markets, then your statistical estimate of covariance is actually ill-defined. So, actually mathematically, you can't use regression on too many markets. And obviously, you don't want to go back, you don't want to use so many observations, you don't want to go back like six months of the CTA index to see what portfolio would best mimic the movements of the index over the last six months, because of course the funds in the index are changing the composition of their portfolios regularly in response to the changes in trends of all the markets that they trade. So, for that reason, you can't trade too many markets because then you run into statistical or mathematical challenges with the modeling, but you also want to use enough markets so that you can make sure that you're spanning all of the different kinds of bets that can take place. So, at the risk of sort of staying too long a little bit on this point, because this is important to understand, if you want equity exposure, you could say, well, we get all the equity exposure that we want with our exposure to the S&P or S&P futures, right? And over the 2008 to 2023 period, the S&P was, first of all, the best performing market in the equity space, but not just in the equity space, I think across all available futures markets, it was the best performing market. And it also was the biggest driver. It was the dog that wagged the tail of all of the other equity markets around the world and all of the other bond markets arguably. But if you go back from sort of 2000 to 2008, then this was a period where emerging markets and foreign markets and foreign currencies and commodities drove the vast majority of what was happening in global market space, and ES or the S&P future did not do very well at all. It really lagged most of the other global equity markets. And so if you tried to proxy equity exposure in the CTA index by just using S&P futures, you ended up really lagging. And so you need enough equity futures to make sure that you can capture periods when foreign markets or emerging markets are really dominating. And other periods when domestic markets are dominating, when foreign bonds are dominating versus domestic bonds, and when different types of commodities within the energy complex and in the metals complex, etc, are dominating for different macroeconomic reasons. And these periods were different classes within the different sectors are dominating and others are lagging. They tend to last a long time. They last five to 10 years often. So you want to be able to represent all the big opportunities that are out there, but then not trace so many markets that you run into statistical challenges, or that you're adding markets that are largely redundant to the markets that you might otherwise trade. Well, I want to give you a chance to respond, Andrew. I do know, and correct me if I'm wrong, I believe, you trade S&P 500, EFee futures and EM futures as the three equity futures that you trade. But before we started recording, you guys already started discussing this topic. And Andrew, I thought you had an interesting point, which was, well, if the US is underperforming EM, can you just gross up the beta? Or if the US is outperforming EM, can you just de-gross the beta and can you get away with less than you think? Is it really in the idiosyncratic returns, or can you just sort of gross up and down the beta? And I don't want to bastardize what you were saying. So I hope that is, and maybe you can step in and sort of clarify where the discussion was going before we started recording. I think if you're going to have a replication, it's a very short term bet. We rebound sportfully as weekly. So all you're betting is on Monday, you can get an accurate read of these long and short positions as to what these guys have been doing, let's say over the past four weeks. You get an accurate read. And you just have to be right enough for the next five days. Now, what determines whether you're right over the next five days? It's not that the S&P 500 does better. S&P 500 is up 256% in the 2010s. EM was up 52%. But that's not what matters. What matters is are you capturing the movements of this portfolio? Part of the problem with the fact that your replication is none of us can think in six or eight or ten dimensions. And so it's not saying these guys have X amount of S&P 500 contracts. It's rather saying that they've identified things that are let's say moving up by X amount across their portfolio. And boy, the S&P 500 is a great way of representing that in combination with lots and lots and lots of different positions. So the problem with factor replication is that it is immune for people to be able to visualize what actually happens down at the modeling level. It's not like this thing broke through its breakout level and therefore I can understand why I would be buying it up. Actually, how the model is actually constructed the portfolios on an optimized basis is it's complicated. So I think there is truth to the idea that you do want some more diversification than four factors. We didn't do a model with four factors. I went to one of the largest family offices in the country and proposed a four factor model. And the guy thought it would probably work, but they were like, it's too weird. We don't know how to talk to the family about this. And so the thing is like our philosophy is we don't have a clue what the future is going to look like. And so you do have to diversify to a certain degree. Because back in 2007 or 2008 after the S&P had gone down over that period of time as NASA had been crushed, I don't know what we would have picked. And so you have to pick it. Now, a general rule of thumb back to Adam's point is that if you're looking at 20 days of data, you're kind of not going to have more than 10 factors. Now, you can do clever things statistically to where you can compress the factors that with PCA analysis. You can do all sorts of statistical jumping through hoops to get to a more accurate result. But again, our philosophy is that we're not going to add an 11th factor unless we see a definitive reason why we should. And after we got past a certain point of factors, whether it was eight or nine or ten, it didn't matter anymore. And then so that our focus is we wanted to be the simplest, most efficient portfolio. Because it's also easier when we were investing in this in our portfolios. I want to be able to look at the screen on Monday morning and see things that are going up and down and have a pretty good intuitive feel as to what my portfolio is doing. Well, so let's talk about that. And I'm jumping around my questions here, but you alluded to it a little bit, which was, and Adam mentioned it as well, which is the flexibility of this return-based replication. Because it is ultimately agnostic to how the managers come up with their signals. It's trying to replicate the returns, not the process. You've sort of alluded to the fact that different markets could be combined to proxy a trade in a market that you're actually not trading, whether it's synthetic or not. There's all these interesting benefits or potential benefits to this top-down replication approach, but it can also make the weights of the system harder to interpret sometimes, right? It might be using some combination of, I don't know, this is just an arbitrary thing, but EM equities and 10-year U.S. Treasury futures, deproxy atrade in JGBs, right? And the weights, you can look at the weights of the portfolio and I can only speak from my experience in the system that we run. It's not always clear. And so I'm curious from your perspective and particularly your experience with clients, how do you communicate that? How do you sort of communicate the idea that you see, hey, everyone should be super short bonds based on the last year of bond returns, but we're actually long to five year right now. It's not easy. People's mindsets are to equate a position with a particular view. And the story value of saying, CTAs think wheat is going to go up is much more compelling than the story value of saying a 10 factor quadratic optimization suggests that this is the aggregate portfolio waiting scheme of it. I think what I try to tell people is that usually at least in my experience, it's getting the big things correct. Because in a sense, some people have been lasted, so we look at 20 days, 40 days and 60 days. Let's say we only looked at 20 days. What if 20 days was a really weird 20 days? And the relationships between two assets had been nothing like that historically. And over the past 20 years, they danced in perfect tandem. The replication model is going to, if it's just looking at that window, it's going to have no idea which to choose. And we discovered this, to show you how long we've been doing this back in 2007, when you would have the Euro stock and the Russell 2000 futures contracts in a different model were an incredibly high correlation. And you get these tiny changes in the underlying data and the model would flip back and forth. Flip back and forth. So we talk about what we do in terms of simplicity, but you have to understand we've been doing this for 15 years. We've been watching these things for 15 years. We have tried everything under the sun. We can tell you every ways in which these things can get fooled and they can get failed. There's a balance between, I think when people have really done a deep dive on our models and how they're constructed and what they're seeking to solve for, it's really sophisticated stuff. And it's built on a decade and a half. So here's an example of where these things can get fooled. If you give it bad data, it will come up with very bad answers. So people have given us portfolios to try to replicate outside of this space that are illiquid. And so when things are illiquid, you see things that looks artificially smooth. How does a replication model think about that? It thinks, okay, I have to find something out there that has been going up like a straight line for the past two months. What is it? What is it? Ah, it's a little bit long, the tenure treasury and short the yen against it in this perfect ratio. Okay? And that's what I mean about that is not a durable answer. So in a sense, it's very similar to what Charlie Munger said. If you have to figure out the two decimal places on a calculator, whether you like the idea or not, it's a really bad idea from a new lesson perspective. If you wanted to be so ten for us was enough. There's no magic to it. Twenty windows works, forty windows, sixty day window works. They all work a little bit better in different times. I think what Adam was talking about trying to find things that happened to work in different market regimes. So our general philosophy was to spread our bets a little bit and try to take a look and take out the pernicious human desire to always optimize. To always hope that that one thing that you found where you turned the dials, you had six over here and four over there and two over there, that's the one where you go with a going forward basis. And our battle hardened experience is having watched a lot of people do this stuff is that's where you run into a wall. I don't want to beat this number of contracts or markets concept to death, but I did receive a question on Twitter that I wanted to bring up particularly in light of the fact that Jerry Parker just launched a new ETF. Congratulations to him. It's great seeing more funds in this space come to market. I think we all agree. The more that we can get a highlight on this market, the more likely that we can get people to begin adopting this space, which we think is all of us agree is the end goal here. And I don't want to mischaracterize Jerry's fund, but I believe he has hundreds of positions. And a lot of what he talks about doing is this idea of outlier hunting. And one of the questions that came up was, and this was for all of us, whether you're trading 10 or 30 markets, is there a concern that trading fewer contracts either reduces the amount of internal diversification of the fund or increases the risk that you're going to miss some outlier fat tail in a market you're not trading. You're not trading the big 80 or big 100 or you're not trading hundreds of markets the way that Jerry is. Adam, I'd love to start with you. How would you address this concern? I think it's a really good example of the point that I was trying to make earlier, where you can have group of markets that are highly correlated. So say, EFEE and S&P and emerging markets and maybe a few of the European forces, DAX and the CAC, etc. If you look at them over time, even over most short periods, they're highly correlated. You look at them and say, well, that's just one bet. From a correlation standpoint, you can make that claim, but even very highly correlated markets can have very, very different means. There was very, very different returns. So I'm not sure whether Jerry is trading individual stocks in his ETF. I know he does that in his mutual fund. So individual stocks are another, again, a good example. Most stocks are highly correlated to the S&P, I think over the long term. The correlation of the average stock, even in the Russell 2000, the S&P is above 0.4. And during periods of crisis, like 2008, that typically runs almost to 0.8. But even though the stocks in the index are highly correlated, they don't all have the same mean. So for example, I think I saw a couple of days ago, somebody published that the seven largest stocks in the S&P are responsible for or up something on the order of 55% this year. And all of the other stocks are up 5%. And so there's a very, very good example, right? If you happen to have those seven stocks in your universe, well, they're obviously very strong. They have had very strong uptrems. And you would likely have capitalized on those moves. Now you might say, well, all stocks are just make up one market. But there's a good example of where that's not true, right? In a similar vein in the early 2000s, ES, EFI and emerging markets were highly correlated, but S&P did very, very poorly relative to EFI and emerging markets, right? So I absolutely think if operationally and within the structure that you can trade more markets that are highly liquid, then you have an opportunity to catch more outlier moves. I mean, the other side of that is if you're trading 150 markets, typically each market is going to have a very small, be allocated to a very small proportion of the total portfolio risk when it has a position on. So it's not like you trade 100 markets and five of those markets have just extremely powerful trends. And you are going to capture all of the opportunity of those five markets. At the margin, it will be helpful, but it's unlikely to be transformative, right? And then there's always just a tradeoff. There's a tradeoff between how frequently you can trade the amount of trading costs. That you're going to incur tax treatment, right? If you're trading securities versus futures gains and losses on futures are treated more favorably from a tax perspective in the US than short-term trading in equity securities. So there's a variety of factors to contemplate. One balance having more bets is theoretically fruitful, but there are statistical and practical realities that probably limit the number that you can trade and that will have a meaningful difference at the margin. I have two comments. And that one is, and you've mentioned a very good point, which is that two assets with a high correlation can have a very different return profiles. So the thing is, top down replication is only a bet on short-term correlation structure of the market, right? If we have asset A that's gone up 10% and asset B is gone up 20%, and you can get exposure to either one of those or both of those with either A or B. As long as that correlation holds true over the course the next week. So the argument is, there's sort of the storytelling exercise that people in the space talk about as though, well, sure, it was correlated a lot over the past month. But this week is the week where the one that you picked, you picked A and A goes down, but B goes up. Like sure, that happens all the time. But sometimes A goes up and B goes down. And even this discussion of outliers is like, I get it, the more you positions, the more opportunity you have to make money on things that have outsized gains, but also the smaller position sizes you have. I think there was one guy on a podcast who talked about replication wouldn't work because he didn't have exposure to some esoteric interest rate market. And how much are you going to have? Are you going to bet the farm on a multiple $1 billion managed futures mutual fund on some of this, of course not. But that's I think where there's this disconnect between storytelling in this business and the way people tell stories in this business, which is, of course, you want the hour later. And I also wish that I'd bought Amazon in 1997 when I wanted to buy it just because I liked books and then held it for the past 20-some years. But whether that has a meaningful impact on returns, if these guys were statistically right, when there was a diversions like that, to get it right 80% of the time, we'd be talking about guys with two sharp ratios that you couldn't replicate. The amazing thing about this space is it's been commoditized over the past 30 years. The skills that the fact that we're doing this on a podcast is reflective of the fact that there are people out there who know how to do this. There's been dissemination of information. And it's still great. That's the incredible part of it. And all we're doing is trying to find different ways of giving access to it. You can do it with 10 factors. You can do it with 20 factors. You can do it top down. You can do it bottom up. The fact that all of these work within a reason is that to me is just a phenomenal thing. So I want to dive into maybe one of the, at least from my experience, one of the more subtle risks of top down before jumping to bottom up and getting both of your feedback, because I know that we'll get some divergence there where Adam, you do bottom up, Andrew, you've researched it and have passed on incorporating it. But one of the things that I saw in the March 2023 bond rally was that the way top down responded was interesting and highlighted a potential unique risk that I love both of you to comment on, which was prior to the bond rally. It seemed like in the short term called the 20 days prior, the variants explained in the Sock Gen CTA index by US Treasuries was pretty small. And so we saw the position in US Treasuries, whether two's five ten or long bonds, was pretty small into the rally. And then as you got this significant bond rally and the Sock Gen CTA index had one of its worst five day returns on record, that top down approach suddenly said, "Oh, actually, it looks like all of these people were massively short bonds." And you had this big bond short open up after the sell off happened because it was looking at this prior data in a lag. And so I think generally people appreciate that there is a lagging component that can be accentuated in these turns. But I think there's another element which is sometimes you might have a position that's just not going anywhere. It's not explaining returns, even though CTAs might have a sizeable short in this case. And all the CTAs were caught offside that then, again, not only is the top down lagging, but it's trying to catch up really quickly at an exact time that the CTAs are taking that position off for vol targeting reasons. It's a really subtle, really nuanced thing, but it's something I watched in real time in that March period. And I would love for both of you to just comment on your experience through that March period, how you saw the systems react. Andrew, maybe you saw something entirely different, but we'll start with you. It was a pretty unique, real live test of these sort of replication systems. The fact that replication is delayed, no question is something that people need to think about with respect to the strategy. But let's just take a step back. The entire manager's space, whether it's top down or bottom up, is looking in a real mirror. In fact, the windshield is blacked out. No one cares about tomorrow in this space. Everybody is looking. And the question is, are you looking back three months, six months, nine months, a year? So what I find sort of ironic about the discussion of looking back, the only question is, are you looking back? Either are you looking back for such a short period of time that you become staggeringly nimble? One of the reasons we didn't do the bottom up approach back in 2015, 2014, 2015, is that it wasn't really settled as to what the window lengths are. What happens in a population like this where you've got a lot of dispersion of strategies, is you see people who do better over time and that other people adopt it. Everybody is constantly replicating everybody else in the asset management business. If somebody's doing something well that's working, their competitors don't say, oh, well, that's their idea. Let's go try to find something completely original and different. So what I think has happened since then, and this is that I think there's been a convergence around basic trend is a longer term phenomenon. Tends to work and tends to be more stable. Back then I think the debate was less settled. But on our side, I don't remember exactly what our positions were, but we felt like we were late in de-risking. We also felt like we were late in November of last year. But when we got to the end of March, we weren't that late actually. When we looked at the constituents of the SoxGen CT index, we were just a little bit behind average. The thing is, when somebody's de-risking, and it works, so remember things, when people have, because the only way most people in the space are going to be materially faster in a moment like that, is because they have some statistical overlay that says, forget about what's happened over the past six months, override it with what's happened over the past 48 hours. Historically, it's 50/50 as to whether they're right or not. Because we've been asked, why don't you do that and stuff? Because again, we're not going to introduce something just because we think we're going to be 50/50 right. What happened in March was that you had the initial drop and then a dawning realization that it could even get worse. Then you had the whole visual banking crisis, then you had credit suites. It was like, you stumbled into a repeller and then you turned on a walkaway and the preller caught up with you and got you again. But there are plenty of times over history where those same guys who de-risked, like when we look at replicating funds that de-risk, you can see them de-risk, but often you wish to hadn't de-risked. So again, it's a 50/50 thing. If we hit a crisis like that where it's a good thing to de-risk and unplug your models basically, then if it keeps getting worse, they'll do a lot better. If on the other hand, as often happens in these things, you get these crazy outside moves these guys de-risk at the bottom and then people realize the world isn't any, it comes bouncing back, then we'll do better. But the point is that no one has figured out a way systematically to tell you what we all want, which we don't have, is a way to de-risk before it happens. We risk on March 12th or whatever it was like the day before. But we haven't figured that one out yet. Obviously, March 2020 and March 2023 are good extreme examples of where the lagged effect of top down is very noticeable, just to play off Andrews metaphor of where all of the trend followers are navigating by looking in the rearview mirror, which is absolutely correct. I think it's useful to extend that metaphor, right? The top down tracking portfolio is navigating by looking in the rearview mirror at the rear view mirror that the trend followers are looking into, right? So at some times it's going to be a bit of a fun house mirror, and those are good examples of where that's true. I mean, look, the reality is for a bunch of reasons, one of which is that it's not just the trend that's affecting positions, but also other factors like changes in risk for the individual markets. And in some cases, changes in covariance is between the markets, depending on the fund. And probably a variety of other factors are also influencing day in day out, the positions of the funds. So there's going to be a lot of noise. Like if you're just using the regular returns of the underlying markets, so you're not, for example, scaling them by volatility in order to track the CTA index. And that brings in, well, if you're going to scale it by volatility, what scaling factor are you going to use? Is it well matched with the scaling factor that's used by, on average, the underlying funds? Are they volatility scaling at all? So there's a number of different things that are all happening there that are adding noise to their top down replication process that depending on how you're modeling it, you're not directly capturing. And so it's not just that there's a delay. It's also just that sometimes you've just got a bad signal and you're not going to be tracking it well. On average, empirically, we see a daily correlation of call at 0.65 to the index and a monthly on the order of call at 0.8 for top down. And that's fairly consistent, whether we're using kind of a slightly smaller universe of 12 or 13 markets or a larger universe of 27 markets. And then our bottom up typically achieves a higher long-term correlation by reverse engineering the process rather than the holdings. But as I said before, they're nicely complementary in general. Yeah. So I want to get into this bottom up. Maybe to put a bow on the top down is you both implement a process that looks at prior returns to try to track the index and build a portfolio. You both agree that process has efficacy. And honestly, even though Adam, you have one approach that uses 27 markets, I know you also have another model that uses just nine. An Andrew, you use 10. So clearly, we all agree. that you can use a smaller set of factors or markets or whatever you want to call it, and get a high degree of efficacy in replicating maybe not perfectly day-to-day, but again, I think capturing these big movements where you do see a high degree of correlation happen across all these markets. What's so interesting, just to speak directly to the point where we have a model that uses nine markets and a model that uses 27 markets, both of those are running top-down replication. The correlation between the replication strategy of the nine markets and the replication strategy of the 27 markets is about 0.7. So just an example here of kind of the amount of noise that you get and the power of an ensemble. The 27 markets definitely not exactly right, the nine markets definitely not exactly right. They're nicely complimentary, they're obviously seeing slightly different things when they're trying to look in the rearview mirror at the rearview mirror, and when you combine the signals together they do better than either one on their own. One thing I was just going to add is that replication began as an idea back in August 2006, when a guy named Andrew Low published a paper on it and to a man and every alligator in hedge funds that was impossible. Here we are 17 years later and it has worked far better than anything else in the liquid oil swirl if you want broad-based exposure to the broad hedge fund industry. People are clinging with a white Nupel group to their myths about what it is about hedge funds. Remember the thing that we're talking about here though with replication is that we're comparing it to an expensive strategy. If you're a mutual fund, it's 170 basis points. It's a lot better than if you last year, if the Sox and CTN was up 20% before fees and expenses, the underlying guys they were up 28% or something. Our thing, we like fat pitches. Our whole point is we'll take noise, we'll take variation over time. When we started doing this in 2016, again, we launched this underlying strategy. We outperformed in 2016. We outperformed in 2017. We outperformed in 2019. We match in 2020. We outperformed in 2021. We outperformed in 2022. We underperformed in the first quarter of this year. We match in the second quarter without performing in the third quarter. And it's daily liquid and now it's in an ETF, et cetera, et cetera. So that is profound. Right? And so whether you're using 10 or 20 should just be an allocator's first stop. Should be why would I not replicate it if I want broad-based exposure to the space? If I believe that DUN capital is going to keep knocking the lights off, I all means, go with DUN capital. They've been doing this for 49 years. Right? The amount of institutional dollars there. If you are invested in PIMCO funds and you want to use PIMCO as a single manager alongside it, of course, you should do that. It's not one or the other, but just the idea that 10 or 20 or 7 or 4 or 3 or 15 or whatever can get you to a similar result in an efficient way is what drives an outcome that's the same. That's great for investors. I think we all agree on that. 100%. The goal of this episode here is to keep you guys in the weeds though. And I'm going to keep dragging you down. So I want to switch a little bit to the bottom-up approach. Right? So we all agree the top-down approach. I think we're all implementing it. We all like it. Where we start to differ as this bottom-up approach. The process-based replication where you try to say, how are these managed futures managers actually doing what they do? Can we create that process? And then instead of trying to sort of look in the rear view of their rear view, maybe as Adam put it, right? You're trying to build a car alongside them guessing how their car is built to a certain degree. I don't know. Maybe I've destroyed that metaphor. Andrew, I know you don't use that approach. Would love to sort of talk through the research you did in the mid-2010s that led you to say, you want to let it's not something we feel comfortable adopting as a means of replication. And the only added color here that I want to add before we go into this discussion is you have primarily focused on what does include other components like carry-in seasonality and relative value trades, whereas Adam has focused more exclusively on the Sockgen trend index, which is almost exclusively trend followers with a little bit of wiggle there. I think there's an important distinction there, but with that, I'll turn it over to you. Get your thoughts on what sort of the research you did on the bottom-up was and why you shied away from it. Sure. So first of all, the trend sub indexes, if we're doing it today, we'd probably do the trend sub index. And we actually we do do that in certain managed accounts. We don't do it in our core product. When we looked at the space back in 2015, there was a wave of bank products called alternative for premium products. And managed futures years ago were called CTAs and they became managed futures. And then around the mid-2010s, as people were trying to think about what they meant by the underlying driving force, they said it's trend. And then banks started to launch all these different what they would say trend falling products. They would come to you and say, if you like merger, don't pay 220 from merger, our guys will give you a cheap daily liquid way of getting exposure to merger, but they did the same thing with trend. Don't buy managed futures, go buy our trend sub swap basically. And when we looked at it, we raised a whole series of questions, one of which was if everyone's calling this the same thing, how come they look so different? So Goldman's swap product was very different from Credit Suises, it's very different for Deutsche Banks, etc. And then because, again, it's not that hard to build a trend falling model. So we said, well, let's look at building it ourselves. And again, what we found was that you could come up with fantastic historical results that had terrific tracking. But we didn't know if it was by chance or not. And again, we were not immersed in the industry where people now talk about it, when did people settle on this being the absolute optimal window length, etc. It felt to us like there was a lot of dispersion in the space. Plus, trend was not cool back then. That back then trend had started to become commoditized, which meant that anybody who had told their investors they were trend, they quickly pivoted to talk and frame themselves as something else to preserve their fee structure. So what we found is when we looked at the models, some of the work rate, some of them didn't. And we felt that any judgment call. And by the way, so what would you do, Dinevli, you said, well, we used them all. That didn't work too well either. So how many markets do you use? Well, in all of these incremental decisions to us, undercut the idea of we wanted the whole space. By replicating from a top down, you're getting fund A in his 17 different ways of approaching this is getting the cat and B and C and D. And as Adam mentioned in the beginning, they're going to evolve what they do over time. And we want to change what we do. We don't want to plant a flag and say 163 days over 27 days is the optimal representation and make it that because we thought that was stupid. We thought we would be if it has to be that close, there's no robustness and no durability to the model. So I think now when we look at it, yeah, I mean, I think people have settled on longer term models. And those are great. And so if we were building everything from scratch today and hadn't done replication and see what it does, we would consider doing it. But again, when you're starting with something that's working. And then the question is, do we find incremental value in adding it to something that we're doing. And again, we're also the guys who want to make it as simple and straightforward as possible. So unless we see a meaningful benefit to it, we're not going to do it. I don't think it hurts it. But it's just not a bet. The more bets that we make like that, the more risk you're taking that Andrew and Matias and Matt. Figure out make a call and where this business is going to go, then just following what the guys are doing, let them make the calls. That's what we're betting on. So Adam, I want you to respond to that in your response. I'm hoping maybe you can talk a little bit about how you develop the bottom up strategy. Because this is where there is a big distinction. Both of you use a top down. But in the paper you wrote, Adam, you ended up coming up with a blend that was something to the effect of 30% top down 70% bottom up when you mix them together to create your replication. So you have a bottom up tilt, whereas Andrew, you're entirely top down. So there is a pretty significant wedge there in terms of how the ultimate process works. So Adam, maybe you can talk a little bit about how you came up with that bottom up approach, maybe ways in which you think you address some of the concerns that Andrew is coming up with. And then how you thought about finding the right blend of the top down the bottom up to solve this problem. Sure. So the goal was to include as many markets and strategies that were feasible to include and that are known in general to be approaches that big CTAs have used over the years. So we started out modeling breakout strategies at different tenors across the 27 liquid markets. And then we did time-stereism and then we did price-to-moving average and then we did double moving average. So now you've got 27 markets. Oh, and across a variety of different trend lengths. So 20-day trend and 40-day trend and 60-day and 90-day etc. all the way out. And with generalized linear models, you can implement something called elastic, natter, rage regression or lasso. And the idea here is you've got a very large number of variables. So short-term time series trend on. WTI crude oil is one of the things that are going to explain the return to the CTA index and intermediate term trend, time series trend. So you've got all these different ways of trading each of these individual markets divided up by the trend length and the way you're defining trend. And you're going to try to use all these little micromoddles on all these markets that we have strong confidence that the broad CTA space trades in order to determine how to wait all of these different micromoddles in order to best match the long term returns of the underlying index. And the reason we took this approach was because I mean, there's only so many ways to slice and dice trends. And we've been in the business replicating or building trend following and other types of quantitative strategies for going on 14, 15 years. So we were bringing some expertise to the table. We didn't want to say, well, we think the best trend strategy is to trade this definition of trend at this average look back. We wanted to say, let's try to figure out or reverse engineer what all the managers in the CTA index are trading. What models are they trading on average? And we use the best practices of data science in order to make sure that we weren't fooling ourselves about our ability to model this process. So we started by taking a segment of the data and trying to model the returns of the CTA index using all of these different micromoddles on that segment of data. And then we applied what we found best explained the CTA index in that subperiod. And then we applied it to a completely different period to see how well the models did on a period that it hadn't been fit to. And we did that using many different ways of slicing and dicing. It's just called K-fold cross validation. And we also did it on a walk forward basis. And we found that the fits are very tight and very stable over time. And actually that was a bit of a surprise for me because I did think as Andrew has alluded to and as the narrative has gone in the CTA space that managers have gone to longer trade lengths or longer holding periods over time. And in fact we didn't observe that. We observed that the micromoddles that were selected to explain CTA returns over any subperiod generalized really well for the entire period. So that was very encouraging. And also the way we defined trend, whether we used very simple time series momentum or moving average crossover breakout really didn't impact much the explanatory power of our bottom-up model, which was another good sign. So we ended up just choosing a simple one. We chose time series momentum and across a variety of different lookbacks over our 27 different markets. And we were able to very effectively mimic the day to day, week to week, and month to month performance of the CTA trend index. And in fact that over the long term did a much better job even out of sample than the top down approach. And that's why we chose to emphasize the bottom-up approach. First of all because we have a much better sense of why the positions that we have are in the portfolio, when we use the bottom-up approach because we already know what the definitions of the models are that we're using to trade the markets in advance. So that's nice that level of transparency for intuition and risk management purposes. Also empirically it did a better job of fitting the index. But again it is complemented by the top down approach. What I like about the top down approach as a complement is that the top down will account for any innovations or changes that might happen in the CTA space in the future. We didn't really observe much of a change in aggregate across all the different managers that are in the index over the 20 odd years we used to fit the models. But they could change going forward. So it's nice to have those top down models in there to adapt. And then also maybe for example the managers are trading synthetic assets like for example trading the yield curve. So where you're long the 10 year treasury in short the two year or long the two year treasury in short the 10 year or maybe your long WTI crude in short brand or something like that. But we do sometimes observe that kind of positioning in the top down portfolio which might validate the potential for those trades in the underlying manager portfolios. So anyways that's the broad thinking that went into that and they end up being very highly complimentary. Though I will say immediately after the paper was published I think the bottom up dramatically underperform the top down for the first three months. So there's a feather in the cap or chalk one up for the top down approach because it was like almost immediately first three months the bottom up struggled the most in its replication as well as an absolute performance. So let me hit on the risks and I'll ask you to sit tight for a sec because Adam I want to hit you on the risks. One of the interesting things that I thought that did come out from your research was that you did find that the trends had been counter to the narrative fairly stable in length over time. Another interesting thing was that and this wasn't sort of top down design but came out organically from the replication that you got about 25% risk in each of the major categories. Equities, bonds, currencies and commodities which aligns with sort of the conversation you have with many CTAs who are trading this space that they end up equal risk waiting across the major categories. But again not in line with the narrative that that was not true pre 2008. The narrative is after 2008 people had to move heavier into financials. People had to move slower trends because of the asset inflow. All of which sort of speaks to the idea of this space of all that is one of the wonderful things about the top down as its agnostics but as a very real world example a qr who is one of the reporting firms in the Soxon trend index has introduced a significant allocation to economic trend models not price trend models. That clearly the question is can price trend models are they correlated enough to pick up those economic trends. Should we think about retraining the bottom up system using some look back up here and I think that's a question that when I talk to people comes up very frequently if you are doing this bottom up how do you think about that risk of process drift in the underlying managers how do you think about that retraining risk in the models. I guess that one's it to me right. Well what gives us comfort is the fact that we didn't observe material drift over the 20 odd years that we examined in our training process. So there's been lots of reasons to have a shift in strategy whether it's to emphasize financials or constrain your equity beta one way or another but we just didn't see that in aggregate the managers in the trend index may decisions that caused the underlying mechanics of their process to change materially over time. There's always going to be innovations that these models are not going to be able to track. A good example also comes from a qr I understand that their trend models also trade long short equity factor baskets. So for example long or short the value basket or the momentum basket or what have you these are going to be very tricky for managers who only trade futures to capture. There may be some dimensions of those bets that are captured by underlying futures. I know for a long time for example the value basket was very commodity heavy. It was mostly miners and energy companies and so there probably was at that point it was a little easier for futures managers to be able to track the vectors of those bets but there's going to be lots of other times where there's just going to be no decent proxy for it. And even a top down approach is not going to be able to meaningfully improve the tracking ability of the replication strategies as these kinds of innovations propagate. But I would echo Andrew's point that you're buying a commoditized factor. You're not really buying an active strategy where in theory at least the alpha is largely coming from the the day-to-day efforts of the people at these hedge fund companies who are extremely well incentivized, well-resourced, highly qualified people who are constantly trying to beat everyone else at the same game. So I think the Replicate Strategy really get people most of the way there for this specific use case, and there's lots of opportunities to off-road a little bit, to find other factors or other managers that you feel might have different kinds of edges. Andrew, I want to come back to you with a little bit more of a general question. One of the things we haven't addressed yet is just the idea of how close should Replication be able to get. I think sometimes when people hear Replication, they think if you have an S&P 500 tracking fund, right? Those things get within basis points. When you talk about Replicating an Index of the constituent managers change over time, there's strategy drift over time, you have no idea what they're holding in any given period of time. What is a reasonable expectation here for how close Replication should be able to get, and maybe we can talk about a couple of different periods, people evaluating this on a monthly quarterly annual basis. It's noisy. Those are the calls with somebody who's doing due diligence and say, "What's the tracking error?" I said, "I have no idea. I don't look at tracking error." Tracking error to what? Tracking error to the theoretical free trading costs returns of some index of 20 hedge funds. They're trading, you talk about AQR, in their flagship fund. I think they've got hundreds of instruments as well now, and they're trading alternative markets and stuff. We are shooting in the same direction. Again, our philosophical approach is that we've got one job, and we are going to be laser-focused on doing one thing, which is, if that index of hedge funds is up 5% net of fees, can we be up 7? Can we be up 8? And we're going to have plenty of periods where we're going to overshoot that by a lot, and there'll be periods like we had in the first quarter here, we're going to undershoot it by a lot. But you just have to believe the basic idea. Because again, why do we want to have as few instruments as possible? Because it's so efficient. No one can front-run us. Try, go ahead, try to front-run us in the S&P 500 Futures contract. Try to front-run us in the 10-year Treasury Futures contract. Our counter-party's hate us. They don't make any money on us. They don't hate us, they like us. But it's sort of a running joke, it's like we're not a really attractive client, but our belief is, our philosophical belief is that with a very small pool, we can get very, very close to their pre-fe and pre-trading costs returns, which gives us a tailwind. In that band, we're going to be underperforming for periods of time, we're going to be outperforming for periods of time. But if you have a big fee advantage, you tend to come out ahead even with noise. The way that I've always thought about it is, if you look at us over a rolling three-year period, let's say, you may go through a period like in the first quarter, but you'll go through periods like in the first half of last year, where we were really meaningfully outperforming. So one of the statistics that Adam mentioned is the correlation of 0.7 for your top-down model. It would be too low for us. If you look at our monthly returns in sepcion, I think we're in like at 0.9. But again, if we've been doing this for a very long time, we have statistical things that are designed purely to try to extract as much information as we can in the best way that we can. So when we talk about including something like bottom up, we're starting with something that works incredibly well. If you just look at us relative to the constituents of the Sox and CTA index since we started, I think that general strategies outperformed 18 of the 20 guys who run back then, not with the simplicity. So the question is, do you really want to add a bottom up model and 10 different instruments and trading costs and our decisions as to all these different things to get from 90 to 92? And so we're always just starting from a very basic thing. Now, what you guys are doing may work better over time. It's very well made. But again, I don't think we're trying to, I think, have a slightly different mandate than how you're describing what you do. Well, just to add a little bit of some numeracy to what you were saying, even with a.9, I think this is a point many people miss. Even with a.9 correlation, if you assume the index has a vol of 10,.9 correlation will still get you an annualized tracking error of 4.5 percent, which I think is a lot bigger than people anticipate necessarily. It's a conversation we have to continuously have with people, which is, again, this is not S&P 500 tracking. This is not you know the exact weights and the exact when things are coming in and out and it's well disclosed and everyone's doing the market on closed trades. This is a statistical approach and there will be noise in the short term. You guys are sophisticated in a way. You understand what the SoxGen CTA index is. You understand the difference between the SoxGen CTA index and the SoxGen CTA trends up index. You know what the trend indicator is. You know about all this other stuff. The reason we did this in an ETF is because the people who benefit the most from this have stocks and bonds and can add this to it. It doesn't matter for them if the SoxGen CTA index is up 7 and we're up 11 or 12 and whether we've overshot on the upside. What matters is we're up 11 or 12 when they need it and their portfolios. And if we go through a period again, a handful of institutional investors are staring at us relative to the underlying indices to make that assessment. And the first quarter of this year, we were having our front foot because we had to make the argument to people that we get it. This is outside of our expectations in a short period of time, but we don't see anything that was suggested is broken. We can't promise you that it's not broken, but we don't see any evidence of it and believe me, we are looking hard and we're looking under every stone. And if we find something, you know us, we will tell you. But that's what I mean about focus on one thing. So there will be plenty of periods of outperformance and underperformance and noise. But that's because it's much simpler and it's inexact, but just happens to work well. Well, I thought what was interesting when you and I were talking earlier this year was despite there's aspects of the process that are similar in aspects of the process that are different. There was replication issues for both of us trying to figure out why. I mean, when I look at the replication approach that we utilize, it was pretty much a good degree of underperformance relative to what our expectation was. And then more or less for the last five months has flatlined and been exactly inline with expectations or replication without changing the process at all. And to your point, you can search and search and search and try to figure out why there was that underperformance. But the unsatisfying answer is it's statistical replication. Just statistically there will be periods both to the upside and the downside of tracking error. It's a question of over the long run. Are you achieving your goal? But you can invest in the 500 stocks in the S&P 500. So tracking error, it's like how do they hedge for industry do? That's a theoretical number. If I had a billion dollars and I had three billion dollars, I'd spread it across 700 funds. This is the return that I would receive. It's just a totology. So the problem in the managed future space is that the categorization of the space is that what you can't actually invest in. And it has its own issues from a construction perspective, from a, you know, a sock gen makes an arbitrary decision. After a non-trend, every time one of those decisions is made, you get farther and farther away from it. So tracking error to me, I've always thought it was a red herring. I think you care about correlations. You want to know that things are working. Even through the first quarter of this year, even though we were underperforming a lot, our correlations remain high. It wasn't that we were picking as Adam would say. We were identifying some big thing out there. It turns out to me when we looked at statistically, it was all the two-year treasury. Literally, we wished we had a one factor model or a 20 factor model. Having a 10 factor model just didn't work. That's tough evidence. That's a good example. Cory, I think you pointed this out. Maybe it was just internally, but just speaking of underperforming and outperforming, but having very strong correlation, right? The bottom up strategies have substantially lagged the top down strategies this year, very substantially since launch. But the correlation of the bottom up strategies to the CTA-Tren index have been substantially higher than the top down strategies, which is consistent with what we observe over the full horizon. But you can have very strong correlation with an index and still have very material underperformance or outperformance. And speaking of underperformance, just like with any other strategy, it is going to be a fool's game to try and choose which replication product is optimal based on observing the live performance. The tracking error to the benchmark and just the relatively low sharp ratio of that single strategy means that there's a massive range of potential outcomes over horizons as long as five or ten years or more. So the population performance expectation for strategy A could dominate strategy B, but strategy B can still outperform strategy A for five, ten, fifteen, twenty years, right? So if you're going to decide to add a CTA or manage future or trend replication sleeve to your portfolio, do it on the basis of connecting with, or resonating with the thinking that went into it, rather than what you observe in the live performance. - So I have one last question for both of you. And this is the out of the weeds top level. This might be the hardest question though, 'cause this is the one I got on Twitter. And this is, how would you explain managed futures to a fifth grader? And then how would you explain replication to a fifth grader? Anyone bold enough to try first? - Do you like surfing? We do too. (laughing) Oh, look, you shouldn't try. Honestly. I think you're trying to explain it to is the 45 year old person who's muddying your managing. And there, I look, we all agree your goal is to help that person to grow that money as much as possible between now and when they retire, when they die, and to be able to sleep as much between now and then. You didn't need managed futures in the 2010s. Stocks and bonds were doing what they were supposed to be doing, the diversification was working. I think in 2030, the guys who look back, and the guys who were building models in 2030 and looking back at the 2020s, it's not gonna be a question is whether they have managed futures or not. 'Cause where else do you find something that is no correlation to both stocks and bonds and positive correlations? Like that's a seismic existential problem for the model portfolio business. The question is, how do you give them a language to explain why it's good for their clients? And that's not about wave detectors versus clusters. That's all professional level stuff. It's a third leg of a stool, let's Eric McGartle from Simplify beautifully said. This should help you to have a nicer experience over the next 10 years. And therefore, and that's why we're doing it because we want to help you as a client. And I think it's not about the gears, it's about the outcome, it's about the experience. And the problem is this whole business, even this is sold around incremental modeling features that the next guy doesn't have. That stuff doesn't matter to the guys who, whether it's gonna be part of their portfolio or not over the next 10 years. How's that for dodging the question? - Expertly done. - I wonder what kind of fifth graders you talk to that use words like existential seismic and correlation? (laughing) That would be maybe Cory is a fifth grader now that I think about it. Yeah, I mean, there's a lot that would be packed into that, they'd have to understand what markets are. I guess, I don't know. If you've been sad over the last half hour, are you likely to be sad in the next five minutes or super happy? Right? Well, you might say, okay, yeah, I'd probably gonna be sad, right? But can you predict how you're gonna feel tomorrow morning or next week? No, right. So it's the same kind of thing in markets, if the market has been happy, well, over the next few days or weeks, probably it's gonna remain happy. If it's been sad, probably it's gonna remain sad. But I have no idea how it's gonna feel a few months from now or got for a bit a few years from now. You know, Andrew, you're gonna have to try that on a fifth grader in a couple of years, right? We'll both try that on our kids or Adam, you can try that on your youngest now. That's right. To be fair, I had a long time to contemplate that while Andrew was answering himself. You got the unfair. But I think the thing we all have to think about, right? Is the messaging in the space is terrible. It is terrible. It is alienating, it is technical, because the people who are in the space are technical by nature. There's a reason they went into the space. Like, I'm the only history major who ended up in the space. And I happened to work with guys, but again, I worked with guys who are more interested in the results and getting to the right answer. We all collectively agree. We want the best hedge fund managers, the best investors that I've known. How do we get to the right answer in the most straightforward and simplest way? Because when I talk to guys who've built and run managed teachers' funds and everybody thinks they're models genius when they launch it. Nobody launches a model thinking it's not gonna work. Nobody writes an academic paper thinking that my academic thesis is gonna be, I look for two years, I couldn't find anything, I hope you still let me graduate. It's everybody looks for things that work really well to nature of this business. And so we're trying to be as humble as we can about it. We're gonna make the best decisions that we can. There's a huge amount of judgment that, as you see in this, that goes into how you end up deciding to do these things. Our philosophy is we wanna get the most right answer that we can in the most efficient, most simple, most straightforward, most explainable way and focus on doing one thing and doing it right. So bottom up replication can work great. And in a sense, I would argue that some funds out there are effectively bottom up replication of ideas that have been widely promulgated in this industry. It's a very kind of hazy line between replication in that scenario and just running your own single manager fund. But again, the key is just we wanna just focus on one thing and do it well and deliver an ETF that people who haven't been invested in this base can use. So maybe I can put a bow on the whole conversation by simply saying with the incredibly controversial statement, diversification is good. We think managed futures introduce beneficial diversification to folks that have traditionally equity and bond heavy portfolios. And we think replication works. And we hope people have really enjoyed this episode. This was a chance to talk to two people who have thought very deeply about this replication concept. Andrew for years and years with his team, Adam with a new foray into the space, but having run hedge fund strategies in the past, trying to think about how he would do it from the hedge fund manager seat. And again, I think what we found is a lot of stuff in common, some small details that are different, but big picture, a huge amount of agreement that this is a space ripe for innovation. And that replication is something that could be an incredibly powerful concept going forward for all investors. Gentlemen, thank you for joining me. I really appreciate it. Thank you so much, Corey. Thanks, Andrew. Thanks, Corey.