Avoiding the Herd: A Different Path to Alpha ft. Jerome Callut

[00:00:07] Jerome

At any point in time you cannot be in a position that would put you out of business. You can control that to the best of our modeling capacities, of course. It’s not a perfect model, but that’s something we like, thinking ahead, what can go wrong? It can be, as you mentioned, a carry reversal, It can be taper tantrum, you know, a strong shock on crude oil after an OPEC meeting or something like that. And each time we can find something that would not let you sleep well at night; we would add it to the risk system and be protected to some degree.

[00:00:50] Intro

Imagine spending an hour with the world’s greatest traders. Imagine learning from their experiences, their successes, and their failures. Imagine no more. Welcome to Top Traders Unplugged, the place where you can learn from the best hedge fund managers in the world. So, you can take your manager, due diligence or investment career to the next level.

Before we begin today’s conversation, remember to keep two things in all the discussion we’ll have about investment performance is about the past and past performance does not guarantee or even infer anything about future performance. Also understand that there’s a significant risk of financial loss with all investment strategies and you need to request and understand the specific risks from the investment manager about their products before you make investment decisions. Here’s your host, veteran hedge fund manager Niels Kaastrup-Larsen.

[00:01:45] Niels

Welcome to another episode in the Open Interest series on Top Traders Unplugged, hosted by Moritz Siebert.

In life as well as in trading, maintaining a spirit of curiosity and open-mindedness is key and this is precisely what the Open Interest series is all about.

Join Moritz as he engages in candid conversations with seasoned professionals from around the globe to uncover their insights, successes and failures, offering you a unique perspective on the investment landscape. So, with no further ado, please enjoy the conversation.

[00:02:21] Moritz

Hey everyone and thank you for tuning in to episode number 11 of the Open Interest series on Top Traders Unplugged. Today I’ll be speaking with Jerome Callut from DCM Systematic, a quantitative hedge fund based in Geneva, Switzerland.

DCM Systematics Diversified Alpha Fund was launched in 2016 and aims to produce returns that are uncorrelated to trend following CTAs by pursuing a different path to Alpha. In fact, the team around Jerome is very much focused on avoiding trend following trades and instead they emphasize strategies which anticipate the flows of other traders, machine learning techniques, and several behavior based models to distinguish themselves from the Soc Gen CTA Index and other industry benchmarks.

Although trend is meant to play no role at DCM Systematic today, Jerome worked on the BlueTrend CTA models at BlueCrest Capital Management between 2008 and 2013. So, he has a deep understanding of the trend following CTA space.

Prior to BlueCrest, Jerome completed a PhD in machine learning from UCL in Belgium, which is one of the oldest universities in Europe. And after BlueCrest, he and his team founded DCM systematic in 2014.

So let me stop the intro here so that we can move over to the more interesting parts of our conversation. Jerome, welcome to Open Interest. It’s great to have you here.

[00:03:47] Jerome

Hi Moritz. Hi everyone. Thanks for having me.

[00:03:50] Moritz

You are more than welcome. Jerome, we had a pre call with you and your team and you kindly shared some presentation materials about your firm with me. On essentially slide number one or two, it says Diversified Alpha uses a multi-strategy non-trend approach to trade liquid futures markets.

So first off, multi-strategy, I think you’ve mentioned you trade more than 25 single systematic models, but obviously we’d like to understand why you explicitly condition these models to be non-trend following models. What role these non-trend models play based in your risk and portfolio allocation framework and so on.

So maybe you give us the background on how it all started because you have a trend following background and then you decided to do the opposite.

[00:04:36] Jerome

Yes, sure. Indeed. I’ve been working together with Anthony on Blue Trend, which is a very famous trend follower program. And I guess we learned really a lot from this experience, certainly from a technical perspective, but we also were exposed to discretionary traders who taught us that markets are not just a time series, but there are supply and demands and all kinds of constraints.

And so, each time you’re trying to build a mathematical model of something, you would essentially make an abstraction or a simplification of a very complex reality where indeed you have many, many drivers at play and you’re trying to capture a few that have some predictive ability, I would say, in the near term or longer term future. And trend following is one technique that is a little bit like universal, that you can apply to many markets.

And what we witnessed is that the CTA landscape was really dominated by these types of approaches. And we had, if you will, the idea that you could perhaps expand to things like relative value trading, or what we call behavioral models. I’ll speak about that a bit later.

But essentially, that having just a condensed view of momentum as a single driver is perhaps not enough, or at least we can go outside of that space. And I guess as quants we were really eager to explore that avenue, and you know, borrow a little bit more from equity market neutral with a more factor based approach and sophisticated risk system. So, when we started researching for Diversified Alpha, our program, we put on ourselves no constraints in terms of what we could or could not do.

But quite quickly we realized that if we were to offer something to investors, we should not overlap too much with trend following programs because that part of the space is well covered by large market participants. We know of course the large usual suspects there and there’s no, I mean if I put myself in the shoes of an investor, I would not necessarily want to take some startup risk to get exposure to something that is widely available.

So, we kind of want it to become that very distinct part of an allocation that would not overlap with let’s call that traditional investment alternative risk premia or trend following, and be really envisaged as a DNA, something that has its own DNA, and that you can allocate without having to think this is, I have already this, from this program, or that, from another program, but more something like orthogonal to many things. And that’s why we wanted to stay remote or away from trend following.

But this is not a statement about us thinking that’s not right or, you know, an approach that people should not do. We think it’s complementary what we do to that. We want to be sitting next to these types of programs in a, let’s say, broader CTA allocation.

[00:08:19] Moritz

Correct. I understand. So it’s also primarily been a business decision, on your part, to distinguish yourselves from the SocGen CTA index, or other industry participants in that space, so that you can provide something that is complementary and has added value in a portfolio context of investors that have already a CTA allocation.

[00:08:40] Jerome

Correct.

[00:08:42] Moritz

If you added like a trend following component to your current setup or your current slate of single models, more than 25 (I think you mentioned you run), it would probably still improve the risk adjusted returns of the portfolio. You would have a higher correlation then, as a result to trend following. But I presume you would have an even more robust return stream overall.

[00:09:09] Jerome

Yeah, I think it’s a good point that CTAs or trend followers have a positive expected return and that you can find the right mix or the optimal balance between us and them in a way. But we think that’s more the role or the mission of the allocator than our own to do that.

What I really like is that people can look at DCM as fulfilling a certain function or role in the portfolio and that we are not doubling down on something they already have. So that’s what prevailed, of course, in the decision.

[00:09:54] Moritz

Now, you’ve mentioned three different styles as like an overall category into which we can essentially put your trading systems.

One is behavioral, which I think is very interesting because what you do there is that you try to anticipate the flows of other traders, at specific points in time, and exploit them. Those could be end of month effects or it could be the footprint left by trend following traders in the markets, things like that. So that’s category number one.

Then the second one is relative value trading or spread trades, where I think you trade VIX futures, but you also trade all sorts of spreads positions across other markets.

And then thirdly, it is a category called macro which, to me, is a little bit tougher to define because macro can mean so many things. But maybe we can go through these three different buckets and give some examples, or you can provide color and context around what they mean. So, let’s start with behavioral. Anticipating the flows and the traits of other traders is what you do there, correct?

[00:11:11] Jerome

Absolutely. So, it really started with a simple question, I guess between myself and Anthony, which was is it easier to forecast the markets directly or to forecast certain market participants that are using certain strategies?

And I guess as a quant you want to find the path of least resistance towards your objective which is, at the end, making positive returns and dealing with the risk. And if you can take that proxy of okay, perhaps there are large market participants that are indeed using well-known approaches. Think of the risk premia, carry, value, momentum, risk parity, or trend following, as you mentioned.

There is some persistence in their positioning because they are managing very large amounts of capital and they can’t, you know, go from long to short in five minutes. It takes them days, and several days, to unwind a large position. So, you ask yourself, can I somehow look at their signals? I know how these strategies are implemented from past experience of myself and Anthony, and can we try to extrapolate that a little bit in the future? Is it possible?

And so, we started looking at various things but something that stuck was really looking at the term structure and typically trying to anticipate carry traders while looking at the slope of term structures on many instruments. And if you look at that, you can see that a market wouldn’t flip from contango to backwardation every other day. It takes, again, weeks to happen typically.

And so, we started to design models like, you know, filtering systems that can gauge where the state of this configuration is and where it’s evolving in a horizon of five to ten days, perhaps. And that’s interesting.

It started to work pretty well on the effects and splitting carry traders and then we expanded that to the other asset classes. And what we saw is that the same principle, perhaps with some adjustment in terms of the risk system, was working pretty well.

So, that’s reassuring when you see one founding principle not just tied to a very narrow asset class, but you can apply it to many. And that’s how it really started for those. And we started really, I think that they were the first models we looked at, and were convinced, and they had nice properties in terms of holding period a little bit faster. You have to be able to give or take twice as fast as the thing you’re trying to anticipate.

So, a holding period between 5 to10 days, sometimes a bit more, and a very low correlation to the CTA indices, like the SocGen you mentioned. So, they were qualified to be included in the program, so to speak. So, those are the first building blocks. And over the years, you know we have an almost nine year track record, now we expanded on those and looked at different behaviors.

Like another is what we call shock impulse models where we would look at what follows a large shock, like a two sigma shock on the commodity markets. What do you see after that? Is there a continuation that you can typically expect? Like is it completely breaking out in the direction of the shock or is it more like an overshooting that is going to be normalized after a week or so. And that’s more what we see, more like overshooting, getting corrected to a level that is more balanced, more nuanced.

[00:15:35] Moritz

So, if we take an example, maybe on one of these two sigma shocks or, you know, in the equity markets, V shaped recoveries, all that type of stuff. Let’s say The S&P 500 has a two sigma down move relative to whatever basis you’re measuring that. It could be with the upcoming election, it could be for all sorts of reasons. So, you have like a short-term breakout to the downside but you’re not really trying to follow the trend.

I think that in your model framework it’s more like, okay, maybe there are how many billions of risk parity and vol control trades sitting there that now need to get derisked because they’re still long. And now two things have happened. A, prices have gone down. So maybe their models hit and exit, but volatility has likely increased as well.

And if they try to target a certain level of vol which they do not want to exceed, then they need to reduce their positions in order to keep their portfolio within that range. That is, I guess what you’re looking to anticipate in terms of a flow. You will probably then expect that there are people coming to hit bids in the markets, selling their positions, and you would be there for a couple of days relatively short term to exploit that flow pressure which that makes on prices.

[00:17:02] Jerome

I think you’re right in the sense it’s opportunistic. Like we don’t have to have the position all the time. There’s triggers and we can enter and exit reasonably fast.

In the specific examples you were taking equities and the S&P 500. You know, there are two things to say. First, we have an implementation of that principle that would try to keep a market neutrality across different indices to not take too much of a directional shock in case of the markets not going in the direction of the model. So, we’re trying to keep a little bit of a market neutrality there.

And second, I would say is that it’s always like the view of one model, but in reality, there are other models that will focus really much more on what happened when there is a volatility spike. That will probably do something different there and engage into a long volatility type of positioning because that could be, kind of, risk-off type of signal coming there. And we want to offer a little bit of tail risk protection. So that could be different ingredients coming into the final S&P 500 positioning.

[00:18:29] Moritz

So, you could also be buying VIX futures, for instance as a result of that signal.

[00:18:34] Jerome

Exactly.

[00:18:26] Moritz

If we take another example, which I think is still timely and on people’s minds, is the carry trade, or the FX carry space that you’re active in, which you just mentioned. So, we had the unwind, or the partial unwind, who knows, where there is that unwind of the Japanese yen carry trade versus the dollar and other currencies in, I think, early August it was. Did you anticipate that happening or is that also a surprise to your models? And then you react to the price action that you see.

[00:19:10] Jerome

That very instance I think we had different results from different anticipation models and the overall result was a little bit difficult for us too, I must say, on that particular trade. But that being said, it’s always very useful to have these kind of negative data points because that’s essentially where you can potentially improve the risk system. And in that case, you mentioned we have 25 plus models, which is a nice design for diversification. But it can happen that, at times, they are in a consensus view on something.

And when that happens you typically, when you ride generate strong returns, but if you’re not, you might be exposed in pretty narrow types of trades and get hit. And so that’s kind of an element that is now on the table for us to research on something we call the overall portfolio overlay, that can look at the positioning of everything all together and see if we are not aligning too much into one direction and potentially cap this exposure. It can be, you know, direct market exposure, but it can be factors like the carry effects for instance, or it can be risk-on overall in the portfolio, things like that.

So, you take a little bit of the pain for sure, but then you come back a few weeks after and you think, is it just something where there’s nothing we can do about it, or can we perhaps see if there’s a mechanism that can be helpful? And I guess you have to stay cold-blooded, like, not trying to do the change because of what happened recently. You didn’t want to go to that suffering again. It has to be balanced and with back and forth discussion with the team for sure.

[00:21:26] Moritz

Yeah. Okay, great.

So that’s the behavioral bucket looking to benefit from essentially the footprints that are left by risk premia strategies, factor strategies. I think we could probably include the QAS, quantitative investment strategies, all these type of strategies which are run by banks where billions of dollars are traded of client money are allocated to. And all of that leaves skid marks in the markets which you can exploit. And some of these footprints seem to be relatively persistent. I tend to be surprised, still, by the end of month effects which I think are observable.

Even though there’s a whole swath of academic papers and white papers around explaining these inefficiencies. But I guess as human beings or large pools of capitals, they just like to run on a monthly schedule. And that’s just like it is the month. And even though you know, and it’s been demonstrated to you that there is an impact, and that you could be improving your execution by trading ahead of month end or after month end, you still do it at month end because that’s how it’s been done for years. And so, it shows up.

[00:22:42] Jerome

I think something also interesting is, you know, we can, to some extent, gauge the positioning of trend followers or carry traders on the markets we are trading, and we can scale that to the AUM of the total CTA industry. And we know, okay, roughly they are in this kind of positioning.

And then you compare that with CFTC data where you have the net speculators and hedges positions and you can have a decent idea of okay, how much do they represent? Is it a lot?

And if it’s really material, and you observe something that perhaps has to imply they would deleverage or unwind the position, you know is going to have a market impact. And that’s where we take the chance of the best opportunities we have is when we know it’s an over traded or concentrated position that will lead to a certain market impact if the position unwinds.

[00:23:53] Moritz

Now moving on to relative value and spread trading, maybe you can enlighten us there with one or two examples and give a bit more background on what exactly you do there.

[00:24:04] Jerome

Sure, we do different kinds of things DID release to enter into some relative value positioning. So, it can be a spread like an instrument versus another, like brand versus WTI or, you know, something like soy oil versus soy meal, or it can be a larger basket with a zero net exposure or at least a controlled net exposure like on FX or grains, for instance.

We have implemented what we call The Fast Mean Reversion Strategy which really borrows from the stat arb type of models where, let’s take the example of FX, you know, if you look at the cross section of the G10 pairs versus the US dollar, and then you look at the daily returns, you can explain up to 80%, 90% of the variance using a three factor model. The first factor would be every currency versus the dollar. The second would be a carry trade type of position. And the third would be, for instance, European countries versus the rest of the world. And you look at the explained variance for maybe 30 years or even more and you see it’s really explaining the bulk of it like 80%, 85%, as I said.

And that’s quite stable. And if you were to do a more statistical principal component analysis, you would, more or less, find the same factors and the same type of explained variance which is a stylized fact we like because in finance having something roughly stationary is rare. I would say you see things are morphing into something else over time. And here is one stylized fact that seems to hold which is always nice.

So, the first reaction as a quant is, okay, let’s focus on these three drivers. Let’s get everything we can get out of them. But okay, then you can do your carry, or trend on the dollar or anything directional on these three. And that would not necessarily qualify for us because you would align with the thing we’re trying to steer away from.

So, instead of that, we have to play with the 10, 15 percent left of unexplained variance which is anything that is more idiosyncratic to a certain currency and that is typically more short lived, I would say. That’s more like 10, 20 days type of mean reversion.

So, to explain a little bit more, what we do is we would neutralize the known factors in the returns of the 10, of the currency pairs, and then we’re left with what we call residual time series where, you know, it’s kind of detrended and mostly oscillating around zeros.

[00:27:21] Moritz

And that gets you, I guess, into kind of like stat arb type of trades where you’re looking to execute a mean reversion trade, or you’re looking for co-integration and stationarity, and trade around that, or there’s all sorts of things that you might be doing there. And I guess, if you don’t do it in FX, one of the examples maybe that is easier to understand, for some listeners or for us in general, is you mentioned WTI vs Brent, which are two highly correlated markets. So, you would fully expect 80% to 85% of the time these two markets are very correlated intraday, end of day, all the time. Every once in a while they decouple, you know, WTI goes negative, Brent doesn’t.

So, is that something where you’re just looking for these two markets to kind of like get out of whack too much and then you would be looking to take the other side, the mean reversion trade, on that spread?

[00:28:18] Jerome

I guess, in that instance it’s a bit different because with the FX pairs you have instruments that are not too highly correlated. So, there is a little bit of room of maneuver for you to deliver some volatility without leveraging the hell out of your positions.

[00:28:40] Moritz

Correct. Which you would have to do in WTI vs Brent.

[00:28:40] Jerome

Which you would have to do there. That’s correct. So, we would not implement directly a stat arb type of approach there. It would more be something derived from the term structure that can have some element of directional exposure which is limited, I would say. You know, we look at RV as something that basically can complement a directional exposure.

But we see, often times, that if you impose a very strict market neutrality, there is some kind of over trading happening, or you have to leverage too much. So we, in some instances, let some constraints, some slack, if you want, to have a little bit of a directional exposure there.

[00:29:36] Moritz

Say you saw the WTI curve in a more pronounced backwardation (I’m just making that up because it’s not the case right now) then Brent, would you therefore kind of say, okay, you’d like to be long some WTI and maybe you will just have a relatively near dated exposure to one of these contracts which are at this very sloped, backwardated part of the curve. And you have maybe no exposure in Brent at all because you know the Brent curve has less slope than the WTI curve.

[00:30:09] Jerome

Yeah, that’s totally possible. But it’s interesting, indeed, to think of that case where you have perhaps a steeper term structure on WTI, in your example, than you have on Brent. So that means that there’s a roll down yield that you can collect there if you will, but less so on Brent in your example.

That’s a great, actually, setup because now you can think of, let’s call that the carry and the spot components of these two return streams. What you can do is almost neutralize the spot return risk using Brent to offset whatever exposure you have to WTI spots and still collect that roll down yield.

So perhaps you would still have an offsetting position on Brent even though there is nothing to collect from the roll down yield just as a risk neutralizer in that case.

[00:31:14] Moritz

I remember from our pre-call that, in that relative value bucket that we’re just talking about, you also trade the VIX futures which I always find very interesting because I guess a majority of traders… Well let me take that back. It is a zero-sum positioning game at the end of the day, at least in terms of futures. But there is a tendency for a lot of people to be shorting VIX contracts and selling volatility as a risk premium.

Is that something that you would also engage in or are you looking really more to have this, as we touched on earlier, when you have these risk-on events to maybe buy VIX futures or you know, be long future spreads on the VIX curve to have an offsetting effect on your portfolio.

[00:31:55] Jerome

So, we would absolutely not engage into a short VIX positioning because part of our DNA, I think, is to provide that diversification during difficult situations, like not only if you look at the 30-year average. Looking back, CTAs have tended to work well when the S&P 500 was in a drawdown.

We Want to be a bit more proactive than that and be specifically good during these times. And I guess we’ve been demonstrating that on a few occasions in our track record.

So, we would completely forbid the system, the program to have a net short positioning on VIX. That’s not something we allow at all. On the contrary, we would like to promote long VIX positioning in a timely way.

And I guess the only grill there is to be long whenever you have a risk of shock and not bleeding out money the rest of the time. Really that’s our objective really. So, we have this kind of RV trading there that, on average, is helping when there’s no shock on volatility.

Perhaps you can implement a market neutral view on VIX where you’re still short at the back, long at the front, and overall, in a beta just sense your market neutral there. But you don’t lose out on average.

And then we have this, to your point, much faster model that would capture a demand for hedging pressure like now volatility getting priced much more expensive at shorter dated maturity, that gets captured as a potential risk of trigger, and then we would enter into a longer VIX positioning overall. So, I guess the two elements would kind of produce what we want there.

Of course we are not right all the time, there are false alarms, etc. And we can see certain models producing no returns for years and years.

But we know their function, we know their role and we know we want to have them. And that’s important. Another element I think that is perhaps interesting to touch on is the risk management. We see that as equally important as the alpha signal generation. And in your example of VIX it’s very true.

I guess you mentioned people who’ve been hit by events like Volmageddon. And what we try to do is to look at risk in a proactive way, like not just turning down the volume after you’ve been hit or doing a realized volatility of the markets kind of adjustment of your position. But more, what could go spectacularly wrong with a certain asset class.

In the case of VIX, we know what it is. After things like Volmageddon, it’s a volatility explosion linked to a reasonably mild drawdown with the S&P 500 in that case. But still, you don’t know exactly what is going to happen, what’s the magnitude, but you can have a good idea of what’s the shape is like. It would hit the front maturities much more than the second, and so on and so forth.

And all you have to do is have a stylized version of that scenario and inject that in a solver that we’ve designed, and say, okay, imagine this can hit us anytime. And so that’s a constraint we had saying if that were to happen, I don’t want to lose more than, let’s say minus two percent on that trade.

And so that’s quite interesting because at any point in time you cannot be in a position that would put you out of business. You can control that to the best of our modeling capacities, of course. It’s not a perfect model, but that’s something we like. Thinking ahead, what can go wrong? It can be, as you mentioned, a carry reversal, It can be taper tantrum, you know, a strong shock on crude oil after an OPEC meeting or something like that. And each time we can find something that would not let you sleep well at night, we would add it to the risk system and be protected to some degree.

[00:36:59] Moritz

I think this is the part, and it’s good that you mentioned that, Jerome, because I had picked that up in your documents as well, because you speak about proactive and reactive risk management and that is a very important consideration for you at DCM Systematic.

So, I guess you just alluded to the proactive part, which is essentially a replay of past scenarios such as Volmageddon, what would happen to the portfolio today, given its current positioning, like a value at risk play or something like that?

[00:37:31] Jerome

Yes, stress testing, if you will, in an active way.

[00:37:35] Moritz

Yeah, there you go. And then the reactive part is obviously exposed, and it cannot be forecast because, well, you react after the fact and you adjust your positions.

So, in the case of Volmagaddon, the incident of the VIX rallying by more than 100% in a single day and kicking out a bunch of the inverse and leveraged ETFs. I think that is the scenario that you are pointing to. Now, in your case, when you’re long at the front end of the VIX curve and maybe you have a short position further dated, that is a scenario that’s not going to kill you, that’s not going to lose you 2%. That’s going to make you a bunch of money.

[00:38:19] Jerome

Correct.

[00:38:20] Moritz

So, it’s more like maybe the inverse of the Volmageddon, or a complete collapse in volatility, or after Volmageddon has happened and when volatility reverses, what could happen to you then? I think this is the riskier part then?

[00:38:36] Jerome

Absolutely. On the day after, I think the 6th of February 2018, we gave back a bit, but the net PNL on that was plus 3% or plus 4%, actually. So, that was a big swing in terms of the return, but still positive. And I don’t know if you would agree with that, but there’s some asymmetry in the price action of VIX. I mean it can jump up very fast, but there’s typically, I would believe, a more gradual fade out of the action.

And the fact that, as you mentioned, we are also managing risk using volatility, that would mean that if you have collected a bunch of money after a shock, you would naturally reduce your position and have the effect of a take profit, so to speak.

That’s something we saw more in 2020, in March that we collected a bunch of profits and then the book kind of got really reduced in exposure for couple of months or so and indeed saved capital, saved the gains, and then we came back and made more. But yeah, I guess this asymmetry in the way it’s taming, it’s probably in our favor.

[00:40:11] Moritz

Maybe there’s an even larger footprint left in the VIX spectrum or in the volatility spectrum than in any other market because there are many traders, including retail traders all around the world exposed to short volatility trades. And if you have these shocks, it becomes very painful. You see the unrealized loss on your screen and it forces a lot of traders out, I guess. So, it has this additional convexity as well, that is potentially exploitable.

Just quickly back to the reactive part. I remember that you have a systematic hedging program that goes on at DCM Systematic. I guess that is the reactive part of your risk management process, right? Where you systematically adjust positions or systematically implement hedging trades to control the risk of the overall portfolio.

[00:41:05] Jerome

That’s more something we are researching right now. So, we have that in parts, let’s say in the different 25 models, they have their way to neutralize certain factors like you know, US dollar carry, or level of the yield curve, things like that. But that’s, up to now, something very local. And the next step, hopefully this year, is to have the full overview of the portfolio with some hedging trades coming up there.

[00:41:37] Moritz

Got it. The one bucket that we overstepped because we (and that’s absolutely fine) we got pulled into the interesting discussion about risk management and Volmageddon is other macro traits, which is the third and final bucket of I think your overall framework. And that is because macro can be pretty much everything and anything depending on the definition. So, what is it for you? What do you do there?

[00:42:03] Jerome

For us it’s mainly two things really. It’s first I would say slower signals that are typically used on fixed income or FX. Thinking of inflation for instance, we use that and with various ways to gauge that, either directly from CPI prints or from real time now-casted version that we designed in house using, for instance, certain equity sectors. But nevertheless, the view would be a longer term one with holding periods of perhaps over a month, up to a couple of months time.

There’s always a tilt or a twist that we put to these models on FX. For instance, models using interest rates and inflation, we would have a risk system that includes a neutralization to, for instance, a risk-off scenario like we witnessed that many times a carry trade in FX would be hurt when there is a shock on the S&P 500 or more broadly on risk on equities. And so, we kind of have this possibility in our risk system to implement a factor which is not based on FX but based on regression techniques. We know how it links to FX and then say okay, please neutralize that for me. I don’t want exposure to that. And that would, in certain cases, make a huge difference, I must say.

So, that’s the first part of what we call macro, and probably what everybody calls macro. And the second is more what we call cross asset models, which is lead/lag relationships between different asset classes. And that’s interesting actually, because if we have a universe of let’s say 75 futures contracts, roughly speaking, and if you now ask yourself, okay, but how many independent degrees of freedom do I really have in that pack, you see it’s more like 20. We’ve done the exercise, I think a couple of weeks ago, and it was 19. So, you see, you have a strong compression factor.

You have less than half of what’s the number of markets you trade. So, at first, look, you think, oh, it’s bad news for diversification because suddenly I have much less things going on. But then you can turn the argument on its head and say, okay, but where does that go? This compression, where does that happen? And you see it’s happening more and more after 2008, we think as the markets got more globalized and people are holding different markets together, and trading them together, and liquidating them together, means that you have aligned flows and more correlations.

So, what we’ve done there is more on trying to say okay, let’s look at and give you a simple example Brent and can we connect Brent to FX pairs or currencies? So, the game is really, okay, let’s try to get the best proxy of Brent using a bunch of currency pairs, using some kind of robust regression techniques, you can get somewhere. And, in some cases, we can get the proxy being correlated to Brent like 0.5, 0.6, which is not too bad. And then you have a new implementation of Brent through currencies and you can reuse. Let’s say you have a great behavioral signal on Brent. When we’ve talked earlier in the conversation, can I take that and trade my basket of FX pairs using that signal?

And I’m not saying it’s working that simple or all the time, but nevertheless we’ve been able to do a few, let’s say connections between different parts of our universe and kind of transfer one predictive signal from one asset class to another one using that kind of mechanism.

[00:46:37] Moritz

You mentioned 75 markets. I presume that everything you do at DCM Systematic is entirely systematic. There’s no discretionary decision making that goes on, on a day-to-day basis, that would interfere with the portfolio.

[00:46:52] Jerome

Yeah.

[00:46:53] Moritz

Maybe in stress situations or something like that. But what role does automation play at your firm? Like is your tech set up completely automated all the way down to the execution step? Or do you run your signals and it produces, you know, a bunch of orders which you then manually send to the market?

[00:47:13] Jerome

I would call that 98% automated in our case. What we typically do is we have a preview that we are running close to the market open and we can have a little bit of time to drill down if we want to understand why are we trading such a position or an order, drill down to the signals and the risk system in case it’s, for instance, larger than usual. It’s always good to have your eyes on that.

But at trade time we have, again, another preview that we have to manually click send to transmit electronically via fixed protocol to the broker. But there’s no way we can modify a trade, or I don’t like this or that, we don’t intervene like that. But there’s always somebody watching the process and have a monitoring of the fields coming back that this is automated but there’s someone responsible for watching the overall process.

[00:48:29] Moritz

It’s not complete autopilot. There is still a pilot?

[00:48:33] Jerome

Absolutely, yeah.

[00:48:36] Moritz

Now, Jerome, you have a PhD in machine learning. I didn’t ask you. Not yet. I do now, what role machine learning and AI models play in your firm, if any.

[00:48:49] Jerome

Yeah, it’s a good point. I’m not the only one with a PhD in machine learning. There are a few others. Anthony in robotics and Antoine, who graduated from EPF at Lausanne, is very aware of all the latest trends like large language models and the deep learning type of stuff. So, we are well versed into that, I would say.

Now to your point, I guess it’s easy to believe that the breakthrough you see in natural language processing, or image processing video are going to translate directly into fantastic new predictive models in finance. I think this is not the case, or at least not from what we can see.

I guess if I look back, there were several attempts from us, at various places and times, where we’ve tried to learn a model from historical data on daily data. And oftentimes what we see is a noisy version of something we know of, like okay, it’s a risk premium, it’s a carry trade, it’s a momentum signal, for instance, that is not as clean as what you would implement yourself knowing exactly what you want. So, that’s one thing that we’ve seen repeatedly, and I think it’s what I said here is not necessarily true if you go higher frequency.

If you have, let’s say, one minute bar data, you have like on 10-year on the S&P 500, you would have 1 million data points. Our attempts that are still in the research phase are much more promising. I would say there’s something to do there.

But if you look at a, let’s say a daily trading system, oftentimes you see something, you know, encoded in a more noisy version, that’s the essential effect you’re looking for. And second is, you know, when you look at the very incredible things we are seeing with AI now, let’s take large language models, it’s trained using trillions of tokens, really like digesting almost the old Internet. And of course they have billions of parameters, but the number of data points is still an order from magnitudes higher than the number of things they have to estimate. And in finance we don’t have so much data. I mean you have to go higher frequency to get more data, but it’s still a much more limited data set.

And I see also two difficulties. One is the weak signal to noise ratio. And if you think, for instance, as a video recommender, like if we share the same taste on certain movies and you’ve seen a new movie that you rate very high, the chance that I will like it is pretty high. The probability that I will enjoy it is high because there’s not a lot of noise. I mean, it’s pretty clean.

But if you look at, let’s take equities, and you see that a company had a bad earnings surprise, so that should translate it into a negative effect on the stock. But if the Fed is cutting rates at the same time, it could be still a positive price action on that stock. And it’s very difficult for a model to disentangle the two effects. It might then believe that a bad surprise is a good thing in a way. So, it’s a blurred picture that I would call weak signal to noise ratio. That’s one.

And the second is lack of stationarity. Another example where AI did wonders is on predicting on how proteins are folding in three dimensions.
That, I think, is a spectacular breakthrough. But if you think about it, the rules on how these proteins are folding were very much the same 100 years ago and perhaps a million years ago.

Whilst if you look again at markets, the participants are changing their behaviors all the time. If there’s something to be arbitraged, it can be there in your data set from 10 years ago, but absolutely no longer valid right now. So that’s something that kind of makes it more difficult to take advantage of this breakthrough, I would say, in a direct way.

[00:53:38] Moritz

Investors change their behaviors, there’s new and different and other market participants entering the space to exploit that behavior, high frequency training, et cetera, et cetera. So, market structure changes and it’s always in flux. Which is why I think it is so difficult for these higher statistical learning techniques, or machine learning techniques, to really decipher the noise from the signal in our space. We will see what happens in the future. But it is nothing… Well, how shall I say that? It is a work in progress.

[00:54:16] Jerome

A work in progress, I guess we need the right type of architecture. And just to your point, I think it’s true that markets are changing, but earlier in the conversation we mentioned the end of months rebalancing, which was there 20, 30 years ago and still is there. So, you can’t really make a generalization of everything. But in general though, it’s more complex than on other types of data, I would say, to use this type of technique.

[00:54:47] Moritz

Yeah, it’s interesting. There are some behavioral things that seem to be very sticky, like the end of month tendency of human beings, because at the end of the day it is human beings that run end of month accounting, and portfolio adjustments, and we run our NAVs on a monthly basis. That’s just what we do.

[00:55:06] Jerome

It is, it is.

[00:55:07] Moritz

You’ve seen in other parts of that, or in that same context, you had a massive liquidity premium associated with the earlier versions of the S&P GSCI Commodity Index providing liquidity to that essentially roll window of business day 5 to 9 in any given month. The same was true for the Dow Jones AIG, then Dow Jones UBS Commodity Index.

But that has maybe not entirely gone away but largely decayed because it’s been presented to investors, and they have moved their exposure to smarter or applications of the same indices which distribute their liquidity and their exposure across the futures curve and no longer provide as clear a target for other people to exploit. So, that is what keeps our jobs and our day-to-day action so interesting because we’re, yeah, working with a moving target.

[00:56:05] Jerome

Moving target it is, absolutely. So, that’s why we stay very focused on research and we see it as a continuous process. At DCM nothing is set in stone for forever. If we have another conversation in a few years, I’m pretty sure that there would be new stuff in or things removed as we have to evolve with the markets. For sure.

[00:56:28] Moritz

Yeah, I’d love that. And maybe then you can talk to me about your systematic hedging program which will then operate like an overlord on the portfolio.

Jerome, Great. Look, we’re getting close to the one-hour mark. I really enjoyed that conversation. Thank you again for coming on and taking the time. It was a real pleasure to have you on.

I think we learned a bunch. It’s been interesting to learn about you, your firm, DCM Systematic, and podcast listeners as usual, we’ll include some of the most important points in today’s show Notes. Should you have any questions, please email us. The email address is info@toptradersunplugged.com. We’ll pick it up and absolutely respond.

So, a big thank you to all of you for listening and until next time on Open Interest, Thanks a lot.

[00:57:17] Jerome

Thanks a lot, Moritz, for having me.

[00:57:19] Ending

Thanks for listening to Top Traders Unplugged. If you feel you learned something of value from today’s episode, the best way to stay updated is to go on over to iTunes and subscribe to the show so that you’ll be sure to get all the new episodes as they’re released. We have some amazing guests lined up for you. And to ensure our show continues to grow, please leave us an honest rating and review in iTunes. It only takes a minute and it’s the best way to show us you love the podcast. We’ll see you next time on Top Traders Unplugged.