[00:00:00]Corey Hoffstein: Look, the reality is, everything in finance is Stacked on the risk free rate. All we should care about is the excess return. Do we think those excess returns have changed because rates are higher or lower? I think most of us would guess, probably not.

Markets find a reasonable clearing price. AQR has some research that suggests actually some strategies like Trend following do better when rates are higher. So there’s an argument that, hey, rates went up. You should actually do more Stacking of managed futures.

​[00:01:48]Corey Hoffstein: We started this podcast?

[00:01:49]Rodrigo Gordillo:Okay, Corey, why don’t you do the honors. I know that you and Adam have a ton to talk about today. So let’s, why don’t we get to the topics?

[00:01:57]Corey Hoffstein: Yeah, this is going to be a great digest episode. So for folks who haven’t been keeping up with all the writing we have been doing, which we are publishing diligently over at Returnstacked.com, you can check out our Insights section. We continue to crank out new articles every single month. There’s a couple of things that we wrote about in the last month. Some published already, some to be published, that we thought would be great to talk about on this episode, because these are questions that come up all the time for us when we talk about Return Stacking generically. And so in this episode, we are going to tackle three different articles we’ve written.

One, which was published earlier this month called Stacking in a Higher Interest Rate Environment. This one comes up, people asking, does Return Stacking make sense when interest rates are high? Doesn’t it imply a very high cost of borrow? Why would I want to use leverage when the cost of borrow is so high?

That comes up all the time. Steven Braun, one of the PMs on our ETFs, wrote a great article about that, recently just published a piece called Return Stacking and Taxes. Another question we get all the time, very hard to answer, because tax recommendation, first of all, none of us are… Second of all, taxes are very personal, so hard to talk about, but wrote a little bit generically about that topic.

And then finally, we’ve obviously done a lot of work on Stacking Trend following, and focused explicitly on Trend replication, have continued over the last year to get a ton of questions about the details of Trend replication. And so, Steven, once again with Adam, did a whole bunch of addendum tests to the paper that you guys wrote, hopefully getting that published in the next month, but wanted to preview some of the different research ideas that we looked at about improving or questioning the design of the replication approach.

And so those are the big three we’re going to hit today and excited to dive in.

[00:04:00]Rodrigo Gordillo: Published now, by the way.

[00:04:01]Corey Hoffstein: All the addendum is live now.

[00:04:03]Rodrigo Gordillo: You go to Investresolve.com.

[00:04:04]Adam Butler: In the, yep, if you go and download the Trend Replication paper, the addendum is at the back and a huge shout out to Steven who did fantastic work on all those articles. Well done. Working in the, so, just to be clear, if you have downloaded the paper previously, you’ll have to download the whole paper again, and it will include the addendum at the back.

[00:04:27]Mike Philbrick: So the addendum is not an individual piece, just so everyone is clear. If you have it already and you haven’t downloaded it recently, the…

[00:04:33]Corey Hoffstein: This is the part of the podcast where I say, don’t worry, there will be a link in the show notes. And then I never add it to the show notes.

[00:04:40]Mike Philbrick: Blame compliance.

[00:04:43]Corey Hoffstein: All right. Let’s dive in, and I want to start with this idea of Stacking in a Higher Interest Rate Environment, and this is a question that comes up all the time. And I think when people start to bridge the gap of Return Stacking, they get the idea ultimately that leverage requires borrowing.

And I think the first sort of connection they draw is often with a mortgage in a house, and they think to themselves, okay, borrowing to buy a house when interest rates were 3%, it feels very different than 7.5, 8% on my mortgage today. Why would I want to do that?

Isn’t it the same? Aren’t you going to have worse returns Stacking in a higher interest rate environment? And I think what this really misses is sort of some of the basic fundamentals of the idea of excess returns, and Adam, I was hoping maybe you could expand a little bit on how borrowing rates are embedded in futures contracts, and how that can show up in different ways in which we’re Stacking.

[00:05:44]Adam Butler: So when you buy a futures contract, you typically don’t pay for the full notional value of the exposure you’re getting. So let’s say you’re going to buy, you want $100.00 exposure to the S&P and you might have to put sort of $5.00 down. And then what effectively happens is the market loans you the other $95.00 to go out and get that exposure. And there is arbitrage between the various players in the market. They compete for the ability to fund that, and they earn a spread on that funding. And there’ve been several studies on the cost of funding for different ways to get leverage in the market. You can get leveraged by borrowing directly, by getting margin in your account. You can do it via swaps, and you can do it via futures. There’s obviously a variety of other more exotic ways, options, et cetera. And concluded, these studies have mostly concluded that futures provide, if not the lowest financing, then very nearly the lowest financing costs of all ways to borrow, right?

So in a Return Stacking context, typically for some, just to sort of simplify it, we own let’s say $1.00 of the S&P 500 as an ETF, or as individual securities. We’ve got sort of cash funded exposure, and then we’re going to add these futures contracts on top, to run a Trend following strategy, or a Carry strategy, and we’re going to use the cash funded equities, or a small cash buffer, to provide the collateral that’s required to buy the leverage exposure in those futures contracts.

[00:07:42]Corey Hoffstein: I think a lot of what gets lost in this conversation is that, all of Finance is really about excess returns, right? If we go back to Finance 101, one of the first things we talk about is the risk free rate. We talk about the equity risk premium. Maybe you might talk about a bond risk premium, but when we’re just, people compare buying equities through futures, like S&P 500 E-minis versus buying cash equities, in theory, in a very efficient market, we should largely be indifferent, right, to buying S&P 500 cash, versus a fully cash backed S&P 500 futures position, right? Those should, if the borrowing rate is the risk free rate, those should basically net us to the same location, right?

[00:08:31]Rodrigo Gordillo: So, so let me show you, uh, I’m going to share my screen here for a second. So what you’re seeing here are three equity lines. The first one here is if you buy, you’re looking at the SPY or the S&P 500 index, right? So that is what you would get if you buy a fully cashed-up ETF from the S&P 500 from 19, it’s far back in 1982 to now.

So this is an extended data series of the S&P 500. And then what you look at when you look at the blue line, that’s what you would get if you’d simply buy the futures contract for the S&P 500, right? So there’s a big gap between the yellow line and the blue line. And the gap should be the difference in financing the position on the S&P 500, right?

So this is, this blue line here is really what you would get if you had to borrow in order to buy the S&P 500 index. And this black line here is cash, right? So what Corey just said was that we should be indifferent between buying the ETF or the fully paid up security, all S&P 500 equities, or buying the futures contract and holding cash. And when we do that, when we actually do, in this case, I’m going to show, the blue line is going to be 100 percent cash and 100 percent Treasury, the S&P futures. You can see that they’re roughly the same, right? There’s some difference around the edges, maybe, you know, doing 100 percent cash stack is not right, because there’s some, the margin requirements, maybe two, two and a half percent, that you need to kind of leave out of cash.

But roughly speaking, there is very little difference between the cash position and futures, plus getting the yield on cash. So the, in essence, when you’re trading futures contracts, you are trading that first blue line that I showed, right? You’re literally trading the excess returns as a managed futures manager, and trying to extract excess return, trying to extract value from that.

But if you’re just looking at even simply the 10-year Treasury or the 30-year Treasury, what you are going to get is the excess returns above cash that you get for taking duration risk, right? So it is an upward sloping line. It doesn’t always, it isn’t always upper sloping, but it is, it has shown through time to provide returns above cash. And so I guess this is when we can get into the discussion as to why that is. Why do we expect that to be the case?

[00:10:59]Adam Butler: I think it’s useful to sort of just pause, and go right back to the most basic thing here. I think this is a really good illustration, Rodrigo. But really the S&P 500 return is a return on cash, plus the return, the excess return that stocks produce above cash, right? And so if you own a, if you want to get your exposure to the S&P 500 by using futures instead of buying an ETF, typically what you’ll have is a bunch of T-bills in your account. You’re going to use those T-bills to collateralize a position in the S&P 500 futures, and then as a year goes by, you will then earn the return on the T-bills plus the return on the change in the S&P 500 futures, right? And so that, that addition should equal pretty well spot on minus a small difference in funding rate, what you would have earned if you would have just cash-purchased the S&P 500 ETF, which effectively is giving you the return on cash plus the equity

[00:12:20]Corey Hoffstein: And again, this is like going back to Finance 101, but I think it’s a really profound thing to internalize. And I think a lot of people forget it, right? When we’re talking about assets like stocks and bonds, we decompose those returns into the risk free rate, plus the excess return or the expected risk premium that we would expect to earn in that asset.

And so when rates go up, we expect that cash component of the return to go up. And then it’s a question of what do we think is going to happen to the excess return component? So let’s take all the fanciness out of this, right? Because at the end of the day, let’s say we were just wanting to lever the S&P. All you’re doing is adding more of that excess return component.

[00:13:06]Mike Philbrick: We have more cowbell, right?

[00:13:07]Corey Hoffstein: And so it’s a question of, really what this boils down to, is as rates go up and down, we know we’re going to get that cash component. Do we think that expected excess return is going to go up and down? Another way of framing that is, do we think interest rate levels are a good market indicator, or predictor, or market timing signal for excess returns and other asset classes. I think there’s a very strong argument, right, that as rates go up and down, the coincidental returns of stocks and bonds should go up and down because of discounted cashflow, but I’m less convinced that the expected returns should necessarily change.

[00:13:51]Adam Butler: Well, yeah, I mean, the expected returns on bonds, nominal return on bonds will change, but the excess return, all things equal should not be the same for stocks.

[00:14:03]Rodrigo Gordillo: Okay, so devil’s advocate, and the thing that everybody’s thinking here is, well, you’re saying that, and we’re seeing something completely opposite, and not for a little while, right? It’s, we’re going on a couple of years now where cash is giving you more than the 30-year, the 10-year. I don’t know. It feels to me like you should not want to own a futures contract versus…

[00:14:26]Corey Hoffstein: Well, a 10-year, a 10-year is a 10 year investment, right? So to measure it over a one year period is not the correct period to measure those expected returns over. First things first. To that point though, then it comes into, okay, if you don’t want to buy the 10-year futures contract, it should also imply you don’t want to buy a 10-year bond.

If you’re not willing to buy the futures contract, which is just the isolated excess return, I don’t know why you would own the bond, which is going to give you the same cash plus excess return experience. You would just own cash. And this is a conversation I have with people all the time and say, oh, I don’t want to Stack bonds. Good. So you’re holding all cash in your portfolio, no bonds. Oh no, we hold bonds. Great. So you have cash plus the future Stacked.

[00:15:13]Adam Butler: Or you could borrow at the 10-year rate to invest in cash.

[00:15:18]Corey Hoffstein: Right. Yeah.

[00:15:19]Adam Butler: Effectively what the quality strategy would suggest, right?

[00:15:21]Corey Hoffstein: So when you piece this together, I think what hopefully becomes clear is that if you are making the claim that Stacking doesn’t make sense because interest rates are higher, you’re effectively making a market timing call, based solely on the level of interest rates.

That said, if you did believe that market timing calls were effective, then it’s a question of, well, what signals could be used? So for example, let’s say you think the embedded interest rate is going to cause a big drag on returns in stocks, bonds, commodities, certain currencies. Well, great, that might lead to a negative Trend and a Stacked Trend following signal could then just short those assets and earn that Carry.

[00:16:01]Rodrigo Gordillo: Yeah. So, by the way, this has been a call that a lot of global macro managers have been making for a while now, right? What is the order of operations here where cash is higher than duration? Any 10-year, 30-year, what’s first going to need to happen is either rates are going to go down, to go back to some semblance of normality, or more likely if rates aren’t going down, which it seems like they’re not, the next step needs to be the long end needs to go up, for what’s, as long end rates go up, bond prices go down. So that’s a market timing call and that’s what should happen, right? That’s what people are saying. And once that happens and yields go up to a certain level, I think the consensus is four and a half plus, that’s when equity started getting hurt, because everything’s more expensive.

Remember companies borrow at the longer rates, so it’s going to hurt them, and then equities are going to go down. So these are timing calls that I’ve been hearing for 24 months, right? It’s just simply very tough to time. And you know, the order of operations are going to be that the question is when it’s going to happen, and whether if you’re an asset allocator, right, are you making those timing calls? Most advisors aren’t. Like, most people still have a big bond position, and they’re doing it because it’s tough to time and long-term you are taking duration risk, which is a, should provide a positive risk premium above cash, right, and if it doesn’t, then we really have to question capitalism.

[00:17:30]Corey Hoffstein: Well, and if the expectation is that, I’m sorry to cut you off, Mike, but really quickly, just right. The way it would work is, if the expectation is that it didn’t, everyone would sell out of that asset, which would drive the yields up, which would then make it more attractive to buy, right? You should, in a reasonably efficient market, find that market clearing price.

[00:17:47]Rodrigo Gordillo: Exactly.

[00:17:49]Mike Philbrick: I think it also goes to the, so in the Stacking paradigm, we’re not Stacking more cowbell on more cowbell, right? So the whole Stacking paradigm is one of, let’s say you want to own the S&P and you’re wanting to add those active diversifiers that may respond differently in the different regimes around interest rates, that might cause price fluctuations, and potentially to the downside in certain markets. And so you’re Stacking that very different return stream.

So what you’re borrowing for too, as well is not necessarily correlated to the underlying that you’re holding, which you know, like, and trust and want to hold for the long run anyway. And it’s just the function of futures that allows you to Stack on top of that, to be prepared for those shifts in regime where, at some interest rate, maybe that will be deleterious to equities, and you’ll have something in your portfolio that’s already built in that kicks in to offset some of those potential negative consequences to that equity component, in this example.

[00:18:55]Rodrigo Gordillo: Exactly.

[00:18:56]Adam Butler: Taxes. Taxes.

[00:18:58]Corey Hoffstein: All right. Let’s jump. So anyone want a quick bow on that? Quick bow is look, the reality is everything in finance is Stacked on the risk free rate. All we should care about is the excess return. Do we think those excess returns have changed because rates are higher or lower? I think most of us would guess probably not.

Markets find a reasonable clearing price. AQR has some research that suggests, actually some strategies like Trend following, do better when rates are higher. So there’s an argument that, hey, rates went up, you should actually do more Stacking of managed futures.

[00:19:33]Rodrigo Gordillo: Yeah. And ultimately look at everything that you’ve heard about in your whole investment career, the equity risk premium, Sharpe ratios, you know, all these things that happen to be things you look for and are positive. These are all X real returns. They’re above cash, right? So if you need to, if you’re quoting equity risk premium, if you’re quoting Sharpe ratios, you’re expecting returns above what it would give you to invest in a simple savings account, okay?

[00:20:04]Rodrigo Gordillo: We kind of all have to buy into that, and then understand that Stacking is anything, really, that has any sort of expected positive risk premium, should do well in all types of interest rate environments.

[00:20:18]Mike Philbrick: The bow for me on this was, I’ll quote the last sentence. In other words, we must believe we can use the level of interest rates as a market timing signal. And it’s not just the level, it’s the expected level. And then it’s the expectations that change around the expected level. That really is hard to measure. And you would have to believe that were the case in order to engage in this type of thing, which is generally not the audience that is going to find Return Stacking appealing, I don’t think.

And then you would have to have a system for which you would execute your belief in the level of interest rates, and the timing methodology that would be used to time your bond and stock investments. That’s really not what Return Stacking is about. Return Stacking is about getting exposure to the betas that you know, love, and trust, and adding a non-correlated source of return on top of that to help you outperform the betas, reduce the downside, and help fund the obligations of the portfolios that this money is invested to provide for. So to me that last sentence really says it all. And you’re, 99 percent of the time, that’s not the case. People are not using interest rates as a market timing tool.

[00:21:38]Rodrigo Gordillo: All right. Let’s talk about taxes.

[00:21:40]Mike Philbrick: All right, taxes, like Adam said. Oh, you gotta love…

[00:21:44]Rodrigo Gordillo: Well, it’s taxes from a Return Stacking perspective.

[00:21:46]Corey Hoffstein: A Return Stacking perspective, right? And this, things get, there’s a couple of things that come up here. The first is there has been this belief for the last 10 or 15 years that ETFs are just able to avoid all taxes, and that is just uniformly not true.

[00:22:05]Rodrigo Gordillo: But it’s uniformly believed.

[00:22:07]Corey Hoffstein: It is very heavily believed. And I think the first thing we have to make clear here is that we’re not going to go into the crazy details, but well-managed ETFs are able to get rid of a lot of realized capital gains in their securities exposure.

So stocks and bonds, or other ETFs, they’re able to get rid of some of it in the options as well. That is not possible with futures, to date. Maybe someone will figure out a way to do it, but with futures, we are unable to use the, create redeem mechanism to push out the low cost basis futures that we want to get rid of, to the market makers and take advantage of the tax code as it’s written, that makes ETFs so efficient, which means if you’re trading something like futures contracts within an ETF, the taxation will be no different than if you were trading them in your own personal account or a mutual fund, or a hedge fund.

[00:23:10]Rodrigo Gordillo: So, Corey, one point of clarity. There is a well managed equity bond…

[00:23:18]Adam Butler: Okay.

[00:23:19]Rodrigo Gordillo: … ETF can manage the taxes. But my understanding, and correct me if I’m wrong, bonds, bond ETFs, don’t have the same advantages as equity based ETFs, with the ability to do the basket swaps.

[00:23:32]Corey Hoffstein: No, they do.

[00:23:33]Rodrigo Gordillo: …have, I’ll, we’ll have to bring @ECONOMPIC on.

[00:23:35]Corey Hoffstein: No, go look at iShares taxable. They haven’t paid taxes and they pay the coupon out, right? The ordinary income has to get paid, but all like, and that’s no different than the S&P dividend has to be paid out. Like, by law, distributions and income have to be paid out at the end of the year in a 1940-Act vehicle, … fund or an ETF, but there are index additions and deletions that happen in something like the Bloomberg AG, and those additions and deletions are tax managed. You can go look at iShares annual…

[00:24:08]Rodrigo Gordillo: Right. But, but…

[00:24:09]Corey Hoffstein:  …would never.

[00:24:09]Rodrigo Gordillo: Those, that are like high yielding ETFs and so on, where people are expecting a vast majority of the returns to come from the yield,  it’s not going to be…

[00:24:17]Corey Hoffstein: Yeah. If you go look at HYG and look at the price return, it’s a straight line down, right, because all the return over time is going to come from…

[00:24:26]Rodrigo Gordillo: That’s an important point, right? Like, there is taxation, there is dividend taxation, there’s yield taxation, and now we got to discuss what type of taxes are involved in futures, which is…

[00:24:38]Corey Hoffstein: The 60/40, right, with futures, there’s really two things that come up. There’s the financial futures where you get 1256 Contract treatment, which is 60% long-term, 40% short-term on realized gains and losses. And then you have your commodities, and commodities if you trade them in an ETF or a mutual fund, you will create a Form K-1, which I can tell you, running an asset management business is absolutely death to a product. And so what every firm does, I don’t know any firm that doesn’t do this, is they create a Cayman entity. You guys, it’s probably just down the street from you guys, you know, a Cayman entity that in which the commodities are traded within, that avoids the Form K-1, but any of the P& L will, actually just the P, the L gets kept at the Cayman blocker level, the profit will bubble up to the ETF in the form of ordinary income.

And so this is important if you’re investing in a Stacked product where what is getting Stacked on top is any sort of futures exposure, a Trend strategy or Carry strategy or systematic macro strategy that is going to generate, hopefully if there’s profit, right, taxable gains in both the form of short-term/long-term capital gains as well from any commodities trading. No different than any other managed futures fund or ETF in the market.

[00:26:05]Rodrigo Gordillo: And one of the things that we’re often asked is, can you give me some level of certainty as to what’s going to be income, what’s going to be capital gain, what’s going to be short-term, long-term. And the truth is that depending on where the P&L came for that year, it’ll be different, right?

If a hundred percent of your P&L came from the commodity sleeve. Then that’s all going to be taxed through the blocker as income. If 100 percent comes from financial assets, currencies, bonds, and equities, then it’s going to be treated as 60/40, right? So it’s going to vary year over year. And the question really is, is there anything you can do about it? And the answer is, not yet.

[00:26:45]Corey Hoffstein: Yeah. Asset…

[00:26:45]Rodrigo Gordillo: Nobody’s figured that out yet.

[00:26:46]Adam Butler: Clarify what you mean by 60/40, because it could be…

[00:26:49]Rodrigo Gordillo: Sixty percent long-term capital gain, forty percent short-term capital gain.

[00:26:53]Corey Hoffstein: … retirement accounts that can avoid those taxes. The second thing that I think is important here is again, this isn’t any different than any managed future strategy.

And in fact, a Stacked strategy may be more tax efficient. Let’s think about what happens in a traditional managed futures Trend strategy. For example, I’m going to give you a dollar and that dollar is going to go into T-bills and then those T-bills are going to be used as collateral for the managed futures strategy, which is going to be managed exactly how we just talked about with the Cayman blocker for the commodities, et cetera, et cetera, versus a Stacked strategy where all you’re really doing is replacing the T-bills with a chunk of equities. You’re going to have to get the equities through, hopefully, a whole bunch of passive equity exposure that can be well tax-managed, plus a little bit of S&P futures. Those S&P futures, 60/40 tax treatment, the long-term chunk of equity, hopefully no taxes. If it’s well-managed in an ETF, you can hopefully avoid paying any taxes other than the dividend distribution, versus, again going back to the original managed futures fund, just holding T-bills, you’re going to get taxed at a federal level in the U.S. on ordinary income, right, which is not a great tax rate for, well, if you’re doing that, if you read the blog posts, we have to assume the highest tax rate. It depends on obviously your tax status.

[00:28:29]Adam Butler: Do you want to just show the blog post? Because it’s kind of well laid out there.

[00:28:56]Corey Hoffstein: A five minute read, where I break down the components that go into the end expected post-tax return, where we say, what is the expected return of U.S. equities from a price return? What’s the expected dividend return? What percent of the portfolio does that make up, and what’s the associated tax rate of those different components. And when you look at these from a pre-tax basis. I just use JP Morgan’s expected return for equities, which I think is 8.2 percent is what they have, and then I use a long-term expected return for managed futures, long-term excess return.

There’s about three to 3.5, I think I used 3.1 in the article. Pre-taxes you have an ex-joint expected return at 11.3. Post taxes that drops to about 9.5. And so it is a meaningful tax drag, about 170 BIPs. But again, if you were to just do this with Treasuries and not have the S&P 500, not only does the total expected return go down, but the tax drag goes up to about 200 bps.

And which is in line, by the way, if you go to Morningstar and you click on any managed futures fund, you can click on, there’s a price tab and scroll down and it will estimate the annual tax drag of that vehicle.

[00:30:24]Rodrigo Gordillo: So, in essence here, when you’re looking at just investing in the S&P 500, it’s 8.2. When you’re investing in a Stacked version, after tax, you are a 9.5. So you’re Stacking 1.3 percent after…

[00:30:40]Adam Butler: Well, hold on, the S&P also is taxed, right? So…

[00:30:43]Corey Hoffstein: Yeah, because you would have the taxes on the 1.6 dividend.

[00:30:48]Rodrigo Gordillo: So, I mean, the reason this is so, the efficiency that’s 75 percent where you’re buying that tax beneficial position in stocks. It’s basically masking the need to tax the cash yield, right? So what I mean by that is if you’re just investing in the S&P 500 with futures, you’re going to have to, you’re going to buy 98 percent T-bills, 2 percent is going to be margin for the S&P and then the S&P-mini, right?

You’re going to be taxed on the whole 90 plus percent on cash, and you’re going to be taxed 60/40 on the S&P. The other way to do it is you buy SPY for 75%. That’s tax sheltered. Now, basically anything that would be cash tax, the T-bill that would be taxed is no longer taxed. And then the 25 percent does get that regular treatment of T-bills plus the E-mini. So you’re, by Stacking and you’re blocking a lot of the tax, the negative…

[00:31:53]Corey Hoffstein: Yeah, I think the post-tax for something like SPY would be, you know, if the pre-tax expected return is 8.2, that after the tax on the dividend, you’re down to like 7.9. So you lose like 30 BIPS to taxes, even if they’re qualified dividends. So, you know, again, we think the way I, the way I communicate this is, absolutely, if you’re trading futures, you want to try to be thoughtful about your asset allocation. But when it comes to Stacking, the way I think about it is, it’s, Mike’s going to hate this. I’m sure he’s going to have a better food analogy, but it’s like having a cake that you add frosting to. Taxes are going to scrape some of that frosting away. But hopefully there’s still plenty of frosting left over, and frosting you wouldn’t have had otherwise.

[00:32:38]Rodrigo Gordillo: Amen. That’s a great way to end that segment. What do we have…

[00:32:45]Corey Hoffstein: So we got this addendum to the Trend Replication research paper that again, you know, Adam spent a ton of time with his team building this out, writing this paper, published it almost a year and a half ago now, right, Adam? So now out of sample, year and a half of experience, tons of questions have come up from people saying, well, did you think of this? Did you think of that? The answer is most of the time Adam did think of it.

We just didn’t explicitly talk about it in the paper. And so, Steven spent a lot of time, Steven Braun on our team, spent a lot of time just sort of going through all these different tests and ideas, and documenting them, and writing out all these different concepts. And so I think there’s like six addendums. Is that right, Adam?

[00:33:33]Adam Butler: Yeah, six or seven, we’re going to go through them, but yeah, if you go download the paper again, then there’s a whole section at the end called The Addendum where we go through all of these and, you know, we’re going to discuss them briefly today, but they are obviously explained in greater detail in the document.

So we definitely would encourage you to go and download it again. But just in general, the idea here was to examine whether, if we made different choices about how to implement the Trend Replication strategy, what impact might we expect those different choices to have? And I think Steven did a great job of examining most of the most important potential dimensions along which we could have made different decisions.

[00:34:24]Mike Philbrick: And I just want to emphasize the point of trade-off. So as you go through that, and as the listeners are listening, note that there is no silver bullet for everything you consider. There is a yin to the yang. There’s no magic that all of a sudden you do this one thing and it all gets better, right? It’s a function of trade-offs and preferences around those trade-offs. And I just want the audience, as they’re listening, to think of that, that all of what we’re going to talk about, through that lens. I think it’s kind of a critical point and I know you’ve mentioned it. I just want to underline it, bold it, and highlight it as we go forward, because they’re just choices and preferences.

[00:35:11]Adam Butler: Yeah, great point. Also, as a reminder, the Trend Replication paper presents a replication method that involves two different approaches. One is a top-down approach, which is like a return-based style analysis. Basically, we’ve got a portfolio of markets. We’ve got different equity markets, bond markets, commodities and currencies, and we are looking back over, let’s say, the past 40 days of returns to the underlying Trend index. And we’re figuring out what portfolio, both long and short, all of these different markets we would hold, in order to best track the returns of the underlying index over that same period. It’s a little bit more complicated than that, but that’s effectively what’s happening in the top-down. And we have two different top-down models. One is a smaller model that just kind of uses the big muscle movements, we often say. So like the major regional equity indices, only Treasuries and the bond complex, and three or four different representative commodities. So, energy commodity, gold, copper, that sort of thing. So small universe.

And then another one, which is, we call it the sort of medium-universe is 27 different markets. And then, so that’s the top-down. It’s sort of 50/50 between those two, the small universe and the medium universe.

The other approach is bottom-up. And that’s where we take our knowledge of how Trend following strategies are technically built and implemented by Trend following managed futures funds, and we build up a strategy based on trying to figure out which of the underlying markets, and what sort of techniques to trade the underlying markets are best representative of the long-term historical returns of the benchmark. So we’ve got the same sort of 27 markets.

Let’s just pick on crude oil for a second. So we’re going to have a very short-term Trend following strategy on crude oil, maybe an intermediate-term Trend following strategy, so maybe like 20 days, maybe 60 days, 90 day Trend following strategy, 180 day, et cetera, up to longer timeframes.

So we’ve got all these different Trend following strategies on all of these different markets, and then we’re figuring out which combinations of strategies and markets best explain the long -erm returns of the benchmark, right? So top-down, bottom-up. And so we’re going to examine both changes in decisions made on how to implement the top-down, and changes made on how to implement the bottom-up strategy, and we’re going to start with changes to the bottom…

[00:38:22]Corey Hoffstein: Adam, before we dive in there really quickly.

[00:38:25]Adam Butler: … on that.

[00:38:25]Corey Hoffstein: I want to do maybe a little bit of table setting because when people hear about replication, I think they often have a misperception as to how close we would expect to replicate something for which we have no idea what the holdings are, or no idea what the exact process is, right?

When we’re looking at something like called the SocGen CTA index or the SocGen CTA Trend index, those are in many ways, a black box to us and we’re doing best efforts to try to replicate what’s coming out. But you know, there’s, I think there’s a misperception as to what sort of tracking error we would expect, and maybe some people don’t even know what tracking error means. So maybe we can just hit pause really quickly before we dive into the addendum, just reset expectations for what do we think is really even feasible here for how close we can get. And that, sort of, I think tells us and informs us about what is just going to be measurement noise in some of this process change versus what’s actually signal.

[00:39:21]Adam Butler: Yeah, well, actually that’s, what’s tracking error, right? That’s probably something we should cover. So tracking error, well, many people I think are familiar with the concept of volatility, right? So you’re going to look at the returns to a stock or an index or something, and you’re going to measure the standard deviation of those returns, and we call that volatility. So tracking error is just the volatility of the difference in returns between a portfolio’s returns and an underlying benchmark’s returns. So basically every day, you take the portfolio returns and the benchmark returns, you subtract the benchmark returns from the portfolio returns, And then you’ve got a series of differences in returns, and you just take the standard deviation of those.

[00:40:11]Corey Hoffstein: Ok, so perfect replication would be there’s no difference. And so there’s no standard deviation.

[00:40:15]Adam Butler: Exactly. Perfect replication. Every day the returns to the replication strategy are exactly the same as the returns to the benchmark index. And then, if we’re able to replicate it very closely, those returns would be small, and the volatility of those returns would be small. If we’re not able to replicate it very closely, then those, the volatility of those returns of tracking here would be larger.

[00:40:41]Corey Hoffstein: So when we look at something like this,  so I’ll just throw this out there. There is a, the most popular managed futures ETF is a replication strategy. It’s been in the market for five plus years now and has had a daily tracking error to its index of an annualized rate of 9%. Now that’s the price. That’s a little unfair because the index settles at, uses NAV. It’s not fair to compare the price of an ETF to NAV, but if you look at the NAV of the ETF versus the NAV of the underlying index, the tracking error I think is 8% annualized, right, if we’re just assuming it’s a normal distribution means, plus or minus 16 percent to the benchmark as well within normal expectation in a given year. And I think that that catches some people off guard, but…

[00:41:33]Adam Butler: No, that’s a really good point. I mean…

[00:41:36]Corey Hoffstein: I was going to say, just for further context, right, I think that is still, and I’d have to go back and check the plots, but still less of a deviation than if you were to pick, ran any manager randomly in the space and compare their deviation versus the benchmark. That tends to be substantially larger. So in terms of hitting down the middle of the fairway and trying to track the index, it’s still closer than if you just chose a random manager.

[00:42:01]Rodrigo Gordillo: That’s a super important point. That’s an absolutely super important point because the reason you want to invest in an ETF index, whether it’s the S&P 500, is so that you don’t have deviation against what the quote-unquote broad market is in the futures space. There is no index that you can invest in as a retail investor.

The best next thing to make sure that you’re not taking idiosyncratic manager risk, which the biggest thing with managed futures, is it’s wild, right? It can be, you can have a manager in no way that lost money. You choose it correctly. And so that index, whether it’s Morningstar or SocGen or Credit Suisse, they’re aggregating a bunch of managers and extracting the beta of Trend. And then Trend replication approaches are trying to be as close as possible to that. But because we’re not investing directly in those underlying funds, we’re going to have some deviation. But as Corey very importantly alluded to, it’s going to be less than randomly picking any single manager. Okay.

That’s why, even though there’s tracking error, Trend replication is still a very good option to be able to get that big muscle movement of Trend. Anyway, good…

[00:43:16]Mike Philbrick: Corey, on those numbers that you elicited through that conversation, is that about an 85 correlation for that particular strategy to the index, do you recall?

[00:43:32]Corey Hoffstein: Yeah. I believe it was somewhere around an 80, 80 to 85 percent correlation to…

[00:43:38]Mike Philbrick: Right. So, so…

[00:43:39]Adam Butler: You measure it daily or monthly or whatever.

[00:43:41]Mike Philbrick: Certainly, but…

[00:43:42]Corey Hoffstein: Yeah. Daily, the range you’re expecting, right? So daily, I think was down to 60 something. Monthly was north of 80.

[00:43:49]Mike Philbrick: So those, and the numbers you were talking about, that 8 percent deviation is on the monthlies.

[00:43:57]Corey Hoffstein: No, that was daily, was it? That was in the, and again, what needs to be specified here is, this is going into the mechanics of an ETF versus say a mutual fund. An ETF, you’re going to have this price that things settle at, at end of day, that is going to trade at a premium or discount to NAV, which is based on the settlement of the futures.

As an extreme example, gold settles at what, 2 p.m. Eastern? It’s going to keep trading for the rest of the day, but you’re, the NAV of that ETF is going to be based on the gold settlement and the NAV of the hedge funds at which inform the index are also going to be based at 2 p.m.. So if you’re comparing a 4 p.m. price versus a 2 p.m. settle, you get some, you can have some days that if there’s meaningful discrepancy, so you have to be careful about how you measure these things. But long story short, it’s just, it is close, but it’s not tight. Like, there is a width here to what’s …

[00:44:49]Mike Philbrick: Yeah. And this is my point. At 0.85 correlation, people think that’s one, and it’s not one. I just want, you know, the 0.85 correlation or thereabouts is exactly what Corey walked through earlier. So if you need to go back and listen to that 8%, 16% deviation on either side of the index, that’s 0.85 correlation. So just so you can kind of, as an advisor or financial advisor/allocator, thinking through what a 0.85 correlation actually means, that’s what it means. And so we would all sort of take 0.85 or 0.8 on face to be very, very similar, without a lot of discrepancy, but as you’ve illustrated, it’s a little bit noisier than what you might expect.

[00:45:37]Corey Hoffstein: So then the question, back to the original conversation, to the points you were making Adam, it was, how can we try to sharpen that pencil? Can we go from 8% tracking error down to seven? We think, based on our numbers, that the combination of the top-down and the bottom-up, over the long run, brings that eight down closer to a seven. Can I be so bold as to say a sub-seven, six and a half? I think that’s been sort of our expectation. Our out-of-sample experience has not been too far off from that. But again, can we go from seven down to five. Is that feasible? Are there ideas and stones we haven’t overturned?

And that’s a lot of what Stephen was looking at in The Addendum research. And again, I think it’s really important research. So Adam, let’s walk through some of the ideas here and explore what the results were.

[00:46:27]Adam Butler: Yeah, that’s good. So there’s a number of different decisions that are made in the construction of the Trend Replication strategy on the bottom-up. Recall that we are trying to identify which underlying markets and trading strategies on those markets are closest to the ones that are being run on average by the funds in the underlying benchmark index. And we just chose a way to define Trend as the exponentially weighted average of returns over different look-back periods, and all the expert exponentially weighting does is, it very slightly gives the returns that are nearest to today a little bit more weight than the returns that are further away from today. And we just wanted to know which combinations of these strategies at different look-backs with different markets, best explain the underlying index. But there are different ways to define Trend, right? We define Trend as, is the exponentially weighted moving average of returns positive, indicating a positive Trend, or negative, indicating a negative Trend? If it is positive, how positive is it? The magnitude of those average returns also makes a difference.

But there’s other ways to do it. For example, we could use a price relative to a moving average of the price. That’s PMAC here, or we could use two moving averages.

We could use a very short-term moving average, and a very long-term moving average, and the distance between the short-term moving average and the long-term moving average at each given day, right? So in this first analysis, we examined whether using a PMAC-style Trend definition or a dual moving average style Trend definition would have made much difference to tracking error. So again, the goal of the replication strategy is not to maximize returns on its face. The goal is to come as close as possible to the returns of the underlying benchmark. So we’re measuring the success of the strategy by whether it minimizes tracking error to the benchmark, right?

So this is showing the rolling one year tracking error of each of these different Trend specifications, when those Trend specifications are used as inputs to our bottom-up replication model. And the quick takeaway is that there is, so bottom-up is the way that we went with in the paper, and the blue and the yellow are just different implementations. And basically the takeaway here is that the way that we used is just as good, no better, no worse, than other common ways of measuring Trend over the long term. Another question we get is, do you guys use binary versions of Trend or continuous versions? A binary Trend signal means that if it is, if a market has a positive sloping return, then it gets assigned a risk weight of one. If it has a negative sloping return, it’s assigned a risk weight of negative one.

[00:50:18]Corey Hoffstein: All in or all out, right?

[00:50:21]Adam Butler: You’re either in or out. You’re either positive or negative, exactly. In your allocation, there’s no like, I’m a little bit positive or I’m really positive. I’m either positive or negative, right? And so examining whether using our way of defining Trend with binary signals versus the way that we actually use it with continuous signals, you can see that again, over time, there’s no meaningful difference in tracking error. Feel free to stop me if you have any questions

[00:50:54]Corey Hoffstein: Well, I think…

[00:50:55]Adam Butler: One of the other inputs to…

[00:50:56]Corey Hoffstein: … worth just pausing there and saying, again, when we look at those graphs, for folks who download the paper or who are watching this video, right. If you’re listening to the audio, apologies, but one of the things is this, these are graphs of rolling one-year realized tracking error to the index. And as Adam pointed out, what we care about with replication is minimizing that tracking error. We will see at any given point in time, sometimes some of these changes had slightly better, sometimes slightly worse tracking error.

Over the full period, they are not statistically different. They often have tracking error, higher or lower in tracking error at the substantially similar times. And so the large point being, I think adding, changing the signal does not improve things. But I think also that suggests that the original signal chosen was not data mined in any way, right? It was fairly arbitrarily chosen that it was going to be an exponentially weighted continuous signal. It happened to be no better or worse than any of the other potential choices that I think are just as reasonable, that many other Trend followers take.

[00:52:00]Rodrigo Gordillo: Yeah. And I think for those listening, taken away, obviously the lines are very close, though there are, if you can pick at any given point in a given month, that there’s a difference of four or 5 percent between one methodology and another.

[00:52:13]Adam Butler: Yeah, there’s random noise differences in the tracking error over time. Exactly. But there’s no systematic reason to say that one is better than the other.

[00:52:26]Mike Philbrick: Yeah. Better to be approximately right than precisely wrong, right? To go data mine the specific signal that provided the best matching, but, out of sample may not do that.

[00:52:41]Adam Butler: So the other input or another important input to these models is, well, if we’re going to be, if we’re observing a positive Trend, well, how much weight do we want to have in this market? And that weighting is determined by the volatility of the market. So given, if two markets have the same Trend signal, one market has substantially higher volatility than the other. Well, the one with higher volatility will have a smaller capital allocation in the portfolio than the market with low volatility. Okay. So that, again, they have the same …. We want them to express the same amount of risk in the portfolio, so we need to measure that volatility in order to determine the weight we place on it. And in the paper, we use a 40 day exponentially weighted moving average, which again, just means that we’re slightly paying more attention to returns nearer to where we are now, than to returns that are further away, and we examined whether using substantially shorter or longer look back horizons for this volatility measurement, made any difference. And in this chart, we actually do see a pretty substantial decay or impact over time, when we use longer and longer lookbacks to estimate market volatility. So, you know, we use 40-day, the 20-day and the 50-day have approximately the same kind of return profile as the 40-day that we used, but once you sort of go into the 100-days, 150-days, 252-days, then we actually get quite a substantial performance drag on the replication index, versus what we implemented. And this is a well-known phenomenon where volatility does tend to auto-correlate over fairly short horizons in the market. And you know, the further out you go in trying to measure or estimate next day’s volatility by looking back at past returns, than if you use very long look-backs to estimate that, then effectively you’re just getting closer and closer to using the long-term average volatility, rather than trying to figure out what the current volatility state is for that market, and adapt to those volatility states over time. You guys want to add to this at all?

[00:55:16]Corey Hoffstein: No, I was going to ask you what the intuition is, because there is a pretty linear decay in performance income versus the 40-day to the 100, 150, 200. But I think you touched upon it nicely. I, again, there are always going to be short-term periods where another choice would have been optimal. I think the point that stands out here for the longer-term vol measurements was in that regional banking crisis, right, where maybe not cutting bond risk would have been more helpful, or maybe it was the long-term vol would have had you in a lower position, but you know, it’s always hard to say which direction it was going into that crisis, but you see relative to what we were, the process we were using, you actually would have done better with longer-term vol estimates, but over the long run, being more reactive to vol, because of that auto-correlation feature you mentioned has proven to be more effective. And I think that’s, you know, again, a lot of intuition and talking to Trend managers and how they actually do this. You know, I’ve never met a Trend manager who is using vol targeting that uses a 252-day measure. It’s just too slow.

[00:56:26]Adam Butler: For sure. The other thing to mention is that these are gross returns, and so you can imagine there’s a bit of a trade-off between using extremely short look-back periods, which may on a gross basis get you slightly higher accuracy, but will also require a lot more trading around that position over time. So there’s kind of a sweet spot between getting the accuracy as strong as possible, while simultaneously minimizing or optimizing turnover. And so that’s kind of where that 40-, 50-day sweet spot comes in.

[00:57:07]Mike Philbrick: What, the Y axis, that is the…

[00:57:12]Adam Butler: Oh yeah.

[00:57:12]Mike Philbrick: … index. Yeah. Could you…

[00:57:13]Adam Butler: This is the cumulative, the difference in cumulative returns over time for the different implementations. So if you look at the dark blue line at the bottom, well, over the full simulation period, using a 252-day look-back would have led to about a, whatever, 30 percent reduction in total return for the strategy over the full horizon.

[00:57:46]Mike Philbrick: And the 20-day add some value, but it…

[00:57:50]Adam Butler: …

[00:57:50]Mike Philbrick: … adds on a gross basis, which is eaten up by the turnover and other things that would be incurred at trying to do something…

[00:58:00]Adam Butler: Yeah, and the other thing is, you’re talking about, let’s say that’s a three or 4 percent improvement over 30 years. You’re talking about a small fraction of a percent per year, right? Like, it actually just ends up being noise. And after transaction costs, it ends up being a drag. So another question was, well, what if we, instead of using exponentially weighted returns, when we’re identifying positive and negative Trends, we just use equal-weighted returns. So the past 200-day average return or past 90-day average return, instead of slightly front weighting it. And just quickly, you know, this shows that effectively this has no impact over time.

[00:58:47]Corey Hoffstein: There were some interesting things that fell out of this though, right? Because, and we’ll get to them in the next graphs. But the reason this question comes up is because a lot of the academic literature uses just equal-weighted prior returns. And so the question was, is there anything special about equal- versus exponentially-weighted?

I have a special love for this first graph because it’s kind of ridiculous how wide the Y axis is, relative, you know, this is sort of like, if you draw anything on a log scale over a long enough timeframe with a thick enough crayon, you end up with something that looks like a straight line like this.

You know, there is some wiggle here, but largely it is statistically indifferent as to whether you use equally-weighted returns or exponentially-weighted. What if we go to the next section that what fell out, and I thought this was really cool as a question of, well, which Trend lookbacks do you end up overweighting depending if you are exponentially-weighted or equally-weighted, right, because if that exponentially-weighted return is already front-weighted does that mean that you will use more of the longer-term Trends? Will the equally-weighted end up weighting more of the short-term Trends when we’re building that bottom-up?

And so I think this graph, Figure 5, to me was just a very cool graph that my intuition… I’m curious about yours, Adam. My intuition here is the regression process did adjust for the difference between exponentially-weighted and equally-weighted. The equally-weighted process ended up using more short-term signals to make up for the fact that the exponentially-weighted was already overweighting those front months.

[01:00:25]Adam Butler: Yeah, exactly. I mean, it’s, this actually is a little bit misleading because for the exponentially-weighted, it’s not a good indication of the actual weighted average return look-back or the half-life of the return look-back on the Trend metric. So it’s kind of hard to see here, but I do agree, Corey.

It’s nice. He sort of pointed out, you see, there’s a little bit of a bar here on the 40- for the equal-weighted, and that’s just sort of it suggesting, and you’re, you also see higher bars at these shorter look-back lengths for the equal-weight to the blue bars, right. And slightly lower blue bars as you go further out in time. And as you say, all this is doing is it’s making adjustments to the model to the equal-weight strategies that basically end up aligning them on a weighted average basis with what we get with the exponentially-weighted average, right? So in fact, the exponentially-weighted average look-back, or sorry, the weighted average look back for the equal-weight Trends was about 204 days. And for the exponentially-weighted Trends, was about 198 days, which, you know, across all these different markets and all these different strategies is statistically completely irrelevant. So it was a neat validation.

[01:01:55]Corey Hoffstein: So once again, the big takeaway for me there is like the choice didn’t matter, right? And what was interesting is, depending on which choice you made, the regression process, the fitting process that was used here, I just adjusted for that choice.

[01:02:12]Adam Butler: Exactly. And then, how did it change from contract to contract? And we just see again that using equal-weighting versus exponential-weighting had almost no impact on the weights across the different markets that were chosen in the bottom-up replication process. And this was actually, I think, your intuition here was important, Corey, I think, because you were the one who sort of first brought up, well, how different are the loadings if you examine the pre-2008 period and the post-2008 period? Because I think there’s a lot of press during the post-2008 period suggesting that Trend managers had lengthened their lookbacks, where there were maybe more short-term traders or the average look-back prior to 2008 was a little bit shorter, and then post-2008 everyone seemed to get just a little bit longer, and that was borne out by this analysis.

So the blue bars are the weights to the different underlying strategies prior to 2008. And you can see that we get a lot more look-back weightings in the sort of 40-, 60-, 90-, 120-days in that pre-2008 period. We get no weightings at all in the post-2008 period below 120-days. And then of course the average, which is just what we had, what we use is the full period, so the pre-2008 and the post-2008 period, in order to create the models. And that of course, just basically the black bars are somewhere between the yellow and the blue bars, which is kind of what you’d expect, and with a little bit more weighting towards the yellow bars, because there’s just more years of data post-2008 than pre-2008, so it just gets higher weighting in the modeling.

[01:04:23]Corey Hoffstein: So if we can

[01:04:23]Rodrigo Gordillo: Just interesting to see there how the blue bar, which is pre-‘08 has the, like the bars are almost the same height, right? It’s almost like 20 percent across the board from what, from 120-day look-back to 150- to 180- to 220- to 260-, it was pretty evenly distributed across that time.

And then post-‘08, it’s 55 percent of the bar goes to 260-. And then prior to that to ‘21 and prior to that to 180- and 180- is like 10%. So the reason this has been a point of discussion is because it was, I would imagine a lot in the post-‘08, it was really tough to trade. You were constantly getting pushed out of trades, right? It was very noisy. They weren’t making a lot of money. And so I think Trend managers ended up pushing further out so that they were trading less and being kicked, you know, just, they gravitated towards the thing that was working.

[01:05:21]Mike Philbrick: Well, I think it’s probably even more insidious than that, Rod. Those who are focusing on shorter-term Trends went bust. Their returns were whipsawed and they don’t exist.

[01:05:32]Corey Hoffstein: There’s that. But I also think, and we’ll get into this in a second. So this aligns with everyone’s intuition. Oh, things slowed down. And I think the expectation is, well, if you’re fitting the full period, which we are, we are now too fast for the way Trends are today. And I want to, what I want to point out is there’s a very subtle aspect of the math here, which is replication at the index level is two parts. It’s the return of the underlying markets times the weights that those markets are being traded. And those weights are going to be dictated by the Trend systems. If those short-term Trend strategies are having no contribution to the overall P&L, it’s going to look like they’re not being traded.

[01:06:16]Rodrigo Gordillo: Interesting. Yeah.

[01:06:17]Corey Hoffstein: So it could just simply be that short-term Trend didn’t do well in the 2010s and therefore had no contribution, and therefore does not show up when you fit over that period.

[01:06:31]Adam Butler: And we’re going to see that show up in the return variance though.

[01:06:33]Corey Hoffstein: It depends, it…

[01:06:34]Adam Butler: …right?

[01:06:34]Corey Hoffstein: … on how, on the relative variance, right, if you’re using a regression-based process though, that’s trying to eliminate extra variables. Like …, right. It’s going to ultimately, right. It might send those towards zero because it’s just not contributing enough variance. It might, right.

But we see this, we’re going to see this in the next section as well, when it comes to markets. And so I think there’s a data, not a data problem, but I think there’s an extra level of analysis that I would love to try to figure out how to look into here, which is, did people actually slow down?

Or for all the people who are saying, look, replicating with slower things fit better, was it because people actually slowed down or was it because the short-term Trend system stopped contributing P&L? They were still being traded, but the P&L sort of flatlined with those systems. And I don’t know what the answers are.

[01:07:29]Adam Butler: Another perspective on this, which gets a little bit, well, look, the incentives are sort of aligned this way, okay. The longer the look-backs on your Trend metrics, the closer you get to earning the drift on the underlying markets, versus earning the Trend-following series, right?

I mean, at the limit, if your Trend length is the same length as your series, you’re only getting the drift. So markets, especially equity markets, there may have been a macro bias, which is that we’ve got now got a Fed put. I believe a little bit more strongly in equities than I did before, and therefore I want to inject a bias, without directly injecting a bias, right? So by lengthening your Trend link, you are injecting more equity and bond beta into the portfolio. And that may have been either an explicit or an implicit goal of the managers.

[01:08:41]Mike Philbrick: Well, you also had a very strong negative Trend in a lot of the markets and the commodity based markets too, that would have profited from the Trend-following system. And it’s, I think to some degree, the light will go on even more when you talk about the actual asset classes and delve into some of the things there, where it might help the light go on for people here. When you dig into this a little bit more, Adam and Corey, because you know, if something’s in the portfolio but it has a zero return, then it’s going to look like it’s not in the portfolio. And that to me is a little bit more easy to grasp than the actual shorter-term Trend-following numbers. And maybe you can…

[01:09:24]Adam Butler: Well…

[01:09:24]Mike Philbrick: … some light on that as well.

[01:09:26]Adam Butler: There’s a good, that is a good point, but it is also hard to tease that out. Like all things equal, … will penalize those markets that are not contributing very much if they’re subsumed by other markets that they’re highly correlated with, but there’s a lot of other stuff that’s going on there, right? So there’s, you can only read so much into these results for a number of reasons, but I think it, the results in general kind of bear out some of the intuitions that we had going into this analysis, right?

[01:10:07]Mike Philbrick: And all of these dynamics are at play. We should be very careful. We have suspicions about certain things. And it’s most likely that all of these dynamics are at play in some proportion that is really hard to figure out.

[01:10:20]Corey Hoffstein: Yeah. I mean, I know my bias was, well, we should be updating the bottom-up model continuously, right? The top-down is going to be adapting, right, because it uses that very short-term look back, but the bottom-up, why would we fit it over a 20-year period? Aren’t these managers changing how they’re trading? Aren’t they changing the markets? And I think there are two things that bear out in this analysis, and we’ll get to sort of, the sort of final analysis in a moment. But I think one of the things that this sort of really highlighted for me is there are potential numerical issues at play here, that if a certain market, certain individual market just isn’t contributing to P&L for five, six, seven years in a row, it could look like, to a replication model, if you’re not careful, that managers aren’t even trading it, right, depending on how much variance it’s contributing.

[01:11:14]Adam Butler: Yeah. So, a good example of that is let’s see, you’ve got five different, you’ve got four different Trend models and you’re trading them all on a market, and this market is gyrating, it’s up and down and you’re trading it as normal, right? But most of the time, the gyrations end up being such that two of the Trend models offset the other two Trend models and your average Trend signal is very near to zero. So you haven’t sort of taken your foot off the gas with this market, you’re still trading it as usual in the portfolio. It’s just that it hasn’t had any persistent Trends or as much in the way of persistent Trends that would give you enough for it to contribute variance to the returns of the benchmark, that you would pick up on with the regression.

[01:12:12]Corey Hoffstein: So it’ll look like the, your managers aren’t trading that market. The bottom-up system would say, we’re not going to put any weight on that market anymore. And then suddenly if that market had a strong Trend, you would be left not trading it. And the same thing happens in the short-term to the top-down system, right?

But I think again, with that bottom-up system, I wasn’t expecting that to come out of, even if we use rolling five- or six-year periods, but you see very different contributions, at least in my opinion, of commodity returns to managed futures in the pre-‘08 versus the post-‘08 period. And those contributions seem to show up as larger weights. And I don’t think, again, talking to a lot of managers, that there was a sudden dramatic shift in commodity exposure, pre-‘08 versus post-‘08. I could be…

[01:13:03]Rodrigo Gordillo: Because then they didn’t shift away from trading commodities just because they were doing poorly.

[01:13:08]Corey Hoffstein: Well, and yes, right. And even though they had been doing so well prior to ‘08, that would have taken a long time for them to shift away if they did.

[01:13:18]Mike Philbrick: But well, and poorly, it’s a matter of Trend. The downtrend in commodities would have been a very potentially profitable moment.

[01:13:27]Rodrigo Gordillo: In 2014, it certainly was, right? So if you weren’t, if you decided not to trade commodities, it would have been very painful.

[01:13:34]Adam Butler: An interesting thing, too, that might also be a bit illustrative. You’ll notice that in the post-2008 period, CL, which is WTI Crude, has a weight, looks like a weight of zero in the portfolio. So this analysis is suggesting that post-2008, Trend followers in general didn’t trade WTI Crude, which is probably one of the most, if not the most liquid contracts traded by pretty well every CTA everywhere.

[01:14:07]Mike Philbrick: And the most important commodity on the globe.

[01:14:09]Rodrigo Gordillo: And in fact, in I think, 2014, 2015, Energies lost a lot, like 75%, something like that. And most CTAs…

[01:14:18]Corey Hoffstein: Yeah, it’s hugely…

[01:14:19]Rodrigo Gordillo: … period. So if they weren’t trading the CL, then it would have showed up in their numbers. And you can just see it…

[01:14:25]Corey Hoffstein: Another one there, Adam, that I want to point out is gold, silver, and copper, right, where you get this huge step up in weights pre- versus post-2008, and that doesn’t make any logical sense. You’re telling me that CTAs doubled their risk weight to copper, gold, and silver over that period? Like that just, again, that to me screams there’s a numerical problem, not an underlying shift in manager behavior.

[01:14:51]Adam Butler: Yes, well, not so much the numerical problem. I don’t, well, I don’t know if that’s quite what you meant, but there’s definitely some, there are other dynamics at play. There are probably correlation dynamics. What I wanted to highlight here with CL is that what happened here is that CL probably became more correlated in the post-2008 period, with, say, CO, which is Brent, and therefore the trading in Brent just subsumed the returns of trades in WTI crude. So you, it looks like it didn’t even bother to trade WTI crude, but in reality, probably trading both. But the P&L from trading both were so similar that the model, just isolating the post-2008 period, did it pick up on it as being a distinct…

[01:15:45]Corey Hoffstein: Nasdaq’s sort of the opposite example there, right? Where you would expect a high correlation with equities. Nasdaq probably was not a very profitable trade, even on the short side in the 2000s. Is it the high vol and then post-2008, man, what a great Trend that was, contributed heavily, looks like a larger part.

So, long story short, I think what my takeaway from this all was, because we can, we could probably talk about this for hours, but my big takeaway was, my expectation was, well, we could look at rolling five- or six-year periods and keep refitting this model. And I’ve changed my view that there are numerical fitting difficulties that make rolling over a short period possibly problematic for a bottom-up approach, and that you do need substantially more history, potentially 20 years, if we think some of these markets may not Trend for years on end, to actually really get a sense of what these managers are doing.

[01:16:45]Adam Butler: Here’s another example, parenthetically, which I think may drive the point home. So as you know, we run a variety of other systematic strategies, and we use more complex models in order to determine the how, how to trade the markets, right? And there was some intuition a few years ago that maybe the market dynamics, in the more recent period, say post-2010 or post-2015, may have changed so sufficiently because of changes in Fed policy, or Fed intervention, or other macro variables, that the data that we were using to train our models from, say, the mid-1970s, in many cases, in most cases from the early 1980s to present, may not be so relevant. And so we actually isolated that period and just decided we were going to train on the 2010-plus period or the 2015-plus period and examine the out-of-sample performance, and what we observed was that in fact the models trained exclusively on the period prior to 2010, actually delivered substantially better out-of-sample performance than models that were trained only on the 2010 to 2018 period. So again, really counterintuitive, and I think, highlights just how noisy the process is in general, and that the general rule is you just want to use the most data that you possibly can in order to run these types of types of models.

[01:18:37]Corey Hoffstein: So the last piece here, by the way, which I think was fascinating was to say, okay, we’ve got the pre- and the post-2008 data, and then we have the full period fit. Which one has better or worse tracking error, and I mean, these lines to me again, noisy, but they are the same.

[01:18:55]Adam Butler: Yep.

[01:18:56]Corey Hoffstein: So for all that effort, you know…

[01:18:59]Rodrigo Gordillo: And you guys put in a lot…

[01:19:00]Corey Hoffstein: Stephen…

[01:19:01]Rodrigo Gordillo: Just to be clear…

[01:19:02]Corey Hoffstein: …of effort.

[01:19:03]Rodrigo Gordillo: …everybody did. I mean, everybody fretted and everybody, you know, was asking all the right questions and there was a lot of back and …

[01:19:10]Adam Butler: Oh, there was enormous brain damage done all around on, you know, how to make this the most robust, resilient methodology to eliminate as many systemic biases as possible. But…

[01:19:22]Corey Hoffstein: You and Andrew did a lot of work to get this code working so that Steven could churn through this stuff. Yeah, absolutely. A ton of brain damage all around.

[01:19:30]Rodrigo Gordillo: And the conclusion is that we didn’t need to do any of it. Being broadly correct was the correct thing to do all…

[01:19:38]Corey Hoffstein: I think that is the life of a quant, which is you spend 99.9% of the time proving that you’re like, you know, it didn’t matter. These choices, they were robust enough at the beginning anyway. But that’s what you’re supposed…

[01:19:51]Adam Butler: Well, yeah, You’re not …

[01:19:52]Corey Hoffstein: … to ask the questions and then go explore.

[01:19:57]Adam Butler: Yeah, these are just…

[01:19:58]Mike Philbrick: Whatever. Hey, tell yourself whatever makes you feel better, Corey.

[01:20:03]Corey Hoffstein: Alright.

[01:20:05]Adam Butler: The tracking here is the same, whether you use the the post-2008 or not. This is…

[01:20:10]Corey Hoffstein: So much more to digest here. And we’re already at an hour and a half.

[01:20:14]Rodrigo Gordillo: And all I hear is…

[01:20:17]Corey Hoffstein: Well, not for me and Adam. This is, me and Adam are waking up right now. We’re…

[01:20:20]Mike Philbrick: Well, there, there’s…

[01:20:21]Adam Butler: That’s right.

[01:20:21]Corey Hoffstein: … up.

[01:20:22]Rodrigo Gordillo: We could go on for hours. I just wonder.

[01:20:24]Corey Hoffstein: Look, there are so, so there’s so much more to go, but some other things we tackled, for example, which markets do you actually need to replicate well, right? And so what, we go through the bottom-up and top-down process and we start dropping markets and say how does the fit change? What, do we actually need equities? And we find some really interesting facts like, well, when you get to just four markets left, lo and behold, you end up with a market in stocks, a market in bonds, a market in currencies and a market in commodities.

So for everyone asking, do we need equities in our replication? Yeah, to replicate the SocGen Trend Index? Almost certainly. We see what sort of tracking error, how tracking error changes when you have 27 markets versus 10 markets versus just four markets, and we see the sort of shape of that curve. There’s all sorts of really interesting addendums explorations packed in here.

[01:21:22]Corey Hoffstein: So Adam, one of the other questions we got a ton was about the selected universe, right? Because people always ask us, well, did you cherry pick the universe? There’s some other replicators out there that have far fewer contracts. We trade more contracts. Our top-down has two versions of, small universe, a big universe.

Like what actually changes when you change the universe. And so we explore this from a couple of angles, and I like this approach that Steven took and he did it for both the top-down and the bottom-up where he’s basically started with the full universe, and then ask the question, which, if I have to drop a contract, which contract has the least impact on tracking error. So you start with 27

[01:22:07]Adam Butler: Historically.

[01:22:08]Corey Hoffstein: Yep. Yep. You start with 27 contracts and then you run all the 26 contract versions of both the top-down and the bottom-up and say, which, if I dropped one, which 26 contract version, had the least impact? And so he had some, he graphically plots this out to show which contracts get dropped first. And there’s a couple of interesting things that fell out of this for me. You know, some of these contracts ended up being more important than I thought. Some ended up being less important than I thought. Would love to get your thoughts.

But the big one that stood out to me was when you start getting to less than 10 contracts. You absolutely maintain one contract in commodities, currencies, bonds, and equities. Like, it really does require exposure to all four asset classes to minimize that tracking error to the SocGen Index. And that’s a question we get all the time. Like, do you have to trade equities? Well, if you’re trying to replicate the Index, almost certainly.

[01:23:07]Adam Butler: I mean, I think there’s a lot of ways you could take this, right? And this is also a noisy analysis, right? Like, another way to sort of take this a step further would be to bootstrap the returns and then find the distribution of these on like, because so much of this is like, these are highly correlated markets.

The differences are extremely small. Which one you choose to take out has an effect then on all of the other regressions you run thereafter. So this is like a highly path-dependent and nonlinear process. I think it does a good job of highlighting that you need, like you said, at least an equity, a bond, a currency, and a commodity contract in the universe to get any kind of replication quality, but I do think it may understate the value of having a lot more, as many different and diverse contracts available to explain the returns of these, of the benchmark over time, while the recent period has been dominated very, by equities, except for like, maybe a couple of big trending periods for commodities.

We have no idea what the future holds, right? So if we were to run this analysis 10 years from now, it may, might look completely different, in terms of which markets had the greatest impact on explaining the performance of the index over time. But we have high confidence it would still show that you need at least one commodity, one currency, one equity, and one bond in the portfolio to have any shot at a high quality replication.

[01:25:07]Corey Hoffstein: And if you, if we scroll down here and show, I think it’s Figure 12, it actually shows again, going back to the point of like, what sort of fit can you get? What we find with both the top-down and the bottom-up replication is as you add more markets, the reduction in tracking error does just sort of peter out, right? Once you get above 10 or 11 or 12 contracts, you don’t get much more of a substantial reduction in tracking error. In fact, increasing, when you have too many contracts, there is the potential that even, it seems to go up at the end. It probably is just statistical noise. There is marginal benefit to your point though, Adam, you don’t know which 10 are the right 10.

And so there is a significant potential benefit in that diversification. The other thing I’ll add is we, at the beginning of this conversation, we talked about what is reasonable tracking error. Eight percent. We look at this graph and we see numbers higher than eight. When we combine the top-down and bottom-up replication, what we find in our numerical analysis is that they are lowly correlated and there is substantial benefit to holding both. That brings the cumulative tracking error substantially down.

[01:26:15]Adam Butler: Yeah, so just to again highlight, this is the 252-day correlation between the top-down, so the combined top-down and the bottom-up returns over time, right? And you can see that they kind of average around 0.4, 0.3, 0.4, but they can be substantially higher or lower over time. But what’s incredible about the, this is pretty wild when you think about it, you have two replication methods, top-down and bottom-up that over time, both do a very good job of explaining the returns to the underlying benchmark, but they only have a 0.3 to 0.4 correlation to one another. And what this means is when you combine them together, they do a very good job of further minimizing the tracking error of the combined strategy relative to the use of either one on its own.

[01:27:18]Rodrigo Gordillo: Yeah, that what’s important, if you go to the beginning of the paper, at some point, you’ll see that one of the things that people need to wrap their minds around when they think about correlation, what they’re envisioning as two equity lines that move in tandem, right? High correlation, two things are moving in tandem, and two things can move in tandem.

So the drift can be very similar. If you compare the SocGen Trend Index against bottom-up against top-down in a chart, you will see that they will look very, very similar. And then you grab that top-down, bottom-up, and you measure their rolling correlations, and they can be as low as 0.3 on any, because of the daily movements, are going to be different because of the different methodologies, but the drift, independent drift are, they look like a Trend Index, right? It is really about how often do we want to be wrong against the Index. And when you go granular, when you’re looking at weeks and months and quarters, it matters, right? We know it matters when you’re investing. This is what people look at. And so we want to minimize the amount of pain as we are extracting that Trend. And the way to do that is if you’re finding two things that are totally correlated to each other that have similar drift, you put that together, you get a better outcome, right? You got a better behavioral outcome.

[01:28:29]Adam Butler: Yep, exactly. And this last chart just demonstrates that when you combine the two methods, that you get lower tracking error than you get from either one of them on its own, right, over the long term, and in general, you get a much lower variability. You don’t see nearly the same kind of spikes you get in tracking error that you get with either of the top-down or bottom-up methods on its own. So again, highlighting the importance of this ensemble method that we use. That’s it.

[01:29:05]Corey Hoffstein: Man.

[01:29:07]Rodrigo Gordillo: Amazing.

[01:29:07]Corey Hoffstein: … went into that,

[01:29:10]Adam Butler: Yeah. Thanks again to Stephen.

[01:29:12]Rodrigo Gordillo: And Andrew. and you, and Corey, and everybody who asked questions. It was, it was a lot of learnings, it was just interesting to be like, I think I figured it out. Corey, it’s this. I’m like, yeah, it’s totally that. Let’s go figure that out. I’m like, no, it’s not that. It was just an amazing experience. It’s a humbling experience when you think you’re so sure, right?

[01:29:33]Corey Hoffstein: It’s humbling. It’s humbling to spend what was months of work to say, we’re not changing anything, right?

[01:29:43]Rodrigo Gordillo: And the original, and the original design was about, again, I’ll say this again, broad, being broadly correct rather than specifically wrong. It’s about, it was about ensembles. It was about humility in the first place. And what we were then trying to say is like, oh, we’re seeing some short-term tracking error. Can we do this specific thing to help improve it? Can we start actually being smarter about it? And the conclusion was like, no, being, you know, as simple as possible, but not simpler, being diversified, having timing, luck, risk it out of the way, having ensembles is almost always in our experience, now, doing this for 20 years, almost always a solution. So that was, it was validating that the original design made sense, even after poking and prodding.

[01:30:30]Corey Hoffstein: Well, with that, I want to say thank you to anyone who made it this far into the episode,

[01:30:34]Rodrigo Gordillo: Yeah. Good.

[01:30:35]Corey Hoffstein: Really, really, yeah, really appreciate the audience here.

[01:30:40]Adam Butler: I want to say thank you to the editors who are going to be able to take this down to something a lot more…

[01:30:44]Rodrigo Gordillo: I think that’s, there’s a, that’s a lofty goal. I don’t think it’s going to happen at all. I think this is going to be an hour and a half podcast.

[01:30:51]Corey Hoffstein: Promise is that the next Get Stacked episode will be chosen by Mike Philbrick. And I promise you, it’ll be a substantially more exciting and enthralling episode, whatever we talk about, but this was, there was so much research that went on this month, as is always happening behind the scenes.

And so it’s, it’s a rare opportunity that we get to, when we do months and months of research, to really bring it out and get to talk about it. And we wanted to use this platform to talk about some of the things that we are doing behind the scenes that you may not see, right?

[01:31:19]Rodrigo Gordillo: Yeah. And to that effect, I think it’s important to talk about what we’re going to probably cover in the next episode is the Carry white paper that ReSolve put out. You know, if you guys haven’t had an opportunity to take a look at that, go to the InvestResolve website, download the white paper. You know, there’s a, there’s a summary paper as well, if you want to take a look at it. So hopefully we’ll go through a deep, as much of a deep dive in that as we have with Trend, and we’ll learn as we go along, we’ll learn to communicate that better as we go along as well. But I think we’re going to, there’s a lot out there already, and I think we are going to dissect that over the coming episodes and looking forward…

[01:31:58]Adam Butler: As a cherry on top, for those of you who’ve lasted this long, for those who don’t know, we have posted the daily and monthly returns from the Trend strategy described in the Trend Replication paper for download. And we’ve also posted the Carry strategy from the Carry paper returns for download, so look for those in the elusive show notes.

[01:32:28]Mike Philbrick: Also add that if you made it this far and you were, I don’t know, confused, cross eyed and whatnot, there’s, we also produced What is Managed Futures Trend Following, which I think is a very insightful piece, that should be kind of put at the top of your list to read as well, just so you have a conceptual idea what that’s based upon, and how it can be so complementary to, traditional stock and bond portfolios.

So, you know, while we definitely took a dip in the complexity pool, that might be a nice refreshing splash in a more simple and straightforward piece that might help you sort of position this with clients and allocators and things like that.

[01:33:11]Corey Hoffstein: I’ll also add, no, just kidding. I’m not adding anything else.

[01:33:15]Rodrigo Gordillo: That…

[01:33:16]Adam Butler: …

[01:33:29]Rodrigo Gordillo: Get Stacked and we’ll chat in about a month.