TTU136: Philip Seager, Head of Absolute Return at Capital Fund Management

[00:00:06] Philip

I think that it was a much faster move. With our convexity product, we’ve managed to capture some of that. Because it was a short-lived move or even like February 2018 as well, that’s really too quick to be caught by most CTAs. So you just have to be careful of that, that there is protection, but it’s a certain type of protection.

[00:00:38] 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 mind. All the discussion we will 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 product before you make investment decisions.

Here’s your host, veteran hedge fund manager, Niels Kaastrup-Larsen.

[00:01:33] Niels

Hey everyone and welcome to another edition of Top Traders Unplugged, where today, Alan Dunne and I are joined by Philip Seager, Head of Absolute Return at Capital Fund Management as part of our mini-series focusing on the one investment strategy that beat everything else in 2022, namely, trend following and manage futures more broadly. First off, Phil, it’s great to have you on the podcast. Thanks so much for joining us today. We really have been looking forward to our conversation. I hope you’re doing well in Paris.

[00:02:00] Philip

All is good. It’s a pleasure to be here. Thanks for the invite.

[00:02:04] Niels

Absolutely. Now, before we dive into all of the various topics that we’re going to cover, Phil, I wanted to set the stage for the conversation and just give the audience a chance to learn a little bit about CFM and its background. So perhaps you can share just a few highlights, maybe, of the type of strategies that you focus on and also where the business stands today as we are in the early part of 2023.

[00:02:30] Philip

Okay, no problem. CFM started in 1991 and we actually started as a trend follower, so we had a CTA product. I joined the firm in 2000. When I joined, my mandate, my first project, was to diversify that futures program into short-term strategies on futures. So you can see early on, the business model was very much to diversify strategies, not because we felt that trend following was not a good strategy, it was just a question of trying to stay ahead of the curve.

About that time, we also diversified into other asset classes. We started trading stocks, that was about 2001, to be able to stat arb. We started trading single stock options, which was about ’04, ’05 about that time, so we had a follow up into other liquid instruments like options on the S&P, options, on treasuries, etc., etc., plus many other liquid instruments.

Today, we are really multi-strat, multi-asset. We have $9.5 billion under management and about $1 billion of that is in two trend following products. One is a dedicated trend following product and one is a convexity, defensive, crisis alpha product.

We have about 60 researchers. To do everything that we do, you need a lot of IT support, so we have about 120 ITs out of a total of 250 people. We’re based primarily in Paris, but we do have some research staff in New York, which is fairly recent, but we’re looking to build that out.

[00:04:42] Niels

That’s wonderful. That’s a great summary. Actually, it’s really interesting because CFM, in that regard, is a little bit different to some of the other managers that we have been speaking to recently.

[00:04:56] Philip

I’ve been listening to some of the other podcasts. I think that some of our viewpoints are potentially quite different to others. Unconventional, perhaps, not controversial, but somewhat unconventional.

[00:05:11] Niels

Yeah. We can’t wait to dive into those differences for sure. As you will also know that Alan and I, we have this list of major topics and we’ll dive in a little bit alternate and see where we go with that. So, as I always do, at this stage, Alan, why don’t I ask you to take it away from here?

[00:05:34] Alan

Thanks, Niels. Phil, it’s interesting, as you say. I can see how CFM can be seen as being quite different. Obviously, you’re running trend following but it’s, I suppose, relatively a small component of the total assets that you’re running. You’re running a number of different quant strategies. So what’s the philosophy around that? We typically want to get a sense of the manager’s philosophy around markets in general and where they feel they can extract their returns in markets.

[00:06:09] Philip

In terms of our philosophy, what I would say is that we believe that inefficiencies do exist, but they are small and they’re very difficult to exploit. We believe, at least for us, that a scientifically rigorous investment process is the best way for us to proceed in extracting alpha. In order to do that, we hire PhD scientists, predominantly physicists. But more recently, in particular, data scientists. Data scientists didn’t really exist when I first started in this business. That’s certainly something that we’ve invested in. We hire PhDs because when you do a PhD, you’re solving a very challenging problem perhaps.

Normally, it’s a unique problem that nobody has solved before and it’s a very humbling experience and these are skills and values relevant to what we’re trying to do. So we have a very collaborative kind of atmosphere. Teamwork is central to what we’re trying to do. I think this is the way that science really progresses. It’s the same here.

Regarding trend following, as I said, our flagship evolved away from trend following, and then in 2013, we launched a dedicated trend following program. The reasons that we did that, firstly, we saw a commoditization or mutual-fundization of this business. But then also, we wrote a paper which was called Two Centuries of Trend Following, you can kind of tell what we did just from the title. The core result of that paper was a P&L of two centuries of data across the four asset classes: equity indices, fixed income, commodities, and FX. The core result was a very, very stable P&L through those two centuries, and that deepened our belief in trend following as a good strategy. It’s a big out-of-sample result. We had genuinely never seen that data so that was a big tick in the box for trend following.

In that paper, we also addressed the recent performance (we wrote it in about 2012 or so, 2013 and it was in that barren period after the global financial crisis), so we addressed that period of underperformance as being consistent with a statistical fluctuation downwards. We dealt with the robustness of trend following to timescales, which is a great thing about strategies. What we’re always looking for, with strategies, is that if they have input parameters, that you see stability to those input parameters, and that is the case for trend following. Obviously, the faster you get, the more you incur costs.

We also addressed the disappearance of short-term trend following, which is something that happened. We actually experienced that because our shorter-term strategies on futures really struggled at the beginning of the 2010s, albeit on a different time scale to what we were looking at in the paper. A third observation, which is a very important observation, is the regression of the SocGen CTA Index and we wrote another paper on that subject.

When you regress this index (which is a great index: 20 of the biggest names in this space) with trend following, then you get 85%, 90% correlation and you get an outperformance and that outperformance is about 4% per year. That 4% essentially covers execution costs and fees. That’s quite a profound observation. Once you remove trend following from the aggregate CTA performance, there is nothing left. So, what that means is on aggregate, CTAs are offering nothing above and beyond trend following. Obviously, there are some good CTAs that outperform and there are some that underperform, but the aggregate is not offering anything more.

So with all that in mind, we launched a dedicated trend follower, and that is not a replicator. What we tried to do is beat the SocGen CTA Index with implementation skills, and we did actually achieve that. Regarding trend following, we’re big believers that it should be a mainstay in institutional portfolios. I remember back in 2019, there was an episode where certain big names in this space publicly took a step back from trend following saying that the trend was dead.

At that time, we remained firmly entrenched in the camp of trend following supporters for many reasons. It really does tick all of the boxes in terms of statistical significance, in terms of out-of-sample performance, in terms of having an effect which is a believable effect, behaviorally (and that’s backed up by intuition and by experimental evidence), and then the diversification with the S&P. It is genuinely uncorrelated.

Even better than that, there’s a hedge in the biggest equity downside moves. That sort of hedge is something. The convexity argument and the crisis alpha argument was a big subject for us, and hopefully, we can talk more about that.

[00:12:29] Alan

Great. An awful lot to unpack there, as people say, but I guess the two centuries of trend following, that’s very interesting and a few different firms did a century or eight centuries or approached it slightly differently. But it is interesting how durable the return series from trend following is. With respect to that research, you touched on how you could see the 2012, ’13 period, it was not unusual in the long-term time horizon, then we had another drawdown 2016 to 2019.

When you look at those periods of underperformance for trend, over 200 years, is it just random? People obviously want to attribute a narrative to good periods and bad periods. I guess the narrative around those was maybe around central banks and maybe around less pronounced moves in markets. Would you buy into that or do you just purely view it as they were on the left tail of the distribution and it is all just a bit random as to when the returns will arrive?

[00:13:49] Philip

People will always look for a narrative and we look for a narrative. We’re not blindered to what is actually happening in the market. We’ve thought to ourselves, as well, that perhaps policymakers are suppressing volatility so maybe you can incorporate that into models, and that models that could tell you when you should be in the trend and when you shouldn’t be in the trend. But we never found anything that really pointed in that direction.

Our view of this is that trend following has modest risk-adjusted returns which are consistent, really, with the level of risk-adjusted returns that you expect from traditional markets. With those levels of Sharpe ratio, you do get extended periods of time where nothing much happens. Trend following, it’s a strategy that exhibits this feature of being positively skewed.

In order to be positively skewed, what that means is that you get accelerations upwards that exist for short periods of time, and then extended periods of time where not much happens. You’re going to get flat periods. So, what we stated in the paper, and we stand by it, is that these flat periods are consistent with the statistics. It’s difficult to come up with a narrative that really explains why such a period could exist.

[00:15:40] Alan

When you look at the 2010s, which as I said, we had two drawdown periods. I don’t know if you can recall the data. Were there similar decades if you go back over the last 200 years? How tough was that period, would you say, in the context of 200 years if you can remember?

[00:16:02] Philip

It’s a very consistent result. The P&L is very consistent with a Sharpe ratio that is unvarying through the 200 years.

[00:16:18] Alan

Is that about 0.5, 0.6? What do you see that as?

[00:16:23] Philip

In the paper, I think the Sharpe ratio was 0.7, but in the paper, it was an academic paper so we didn’t have costs and we didn’t have fees. Our estimate of trend following provided the clients, net of everything, is a Sharpe ratio of 0.5. That’s what we see consistently through time. Plus, when you look at hedge fund databases like self-proclaimed CTAs and managed futures funds, what you see is that if you look at the distribution of Sharpe ratio versus the amount of time that the fund has been in business, you see a convergence to a Sharpe ratio of 0.5.

That Sharpe ratio of 0.5, it’s not like high octane stuff. Just to put things in perspective, if you look at the equity risk premium, it’s, I would say, probably somewhat lower than that. It’s probably 0.3, 0.4. So a Sharpe ratio of 0.5, in the great scheme of things, is actually pretty good.

[00:17:37] Alan

Yeah, just one final one on this paper, because I think it is very powerful. But I guess people may be skeptical, a little bit. Two hundred centuries ago, markets were… Two centuries not 200. Two hundred centuries would be even more remarkable. Markets were very different. We didn’t have all the futures markets. Obviously, you would have had to do some work on the data. So maybe tell us a little bit about what kind of adjustments you had to make and how realistic do you think the data is going back that far in time?

[00:18:13] Philip

Well, clearly, it’s not realistic that we could have run a business over that 200-year period. That’s not realistic. What is realistic, I think, is that it’s really a question of looking for evidence; this weak effect that price returns are autocorrelated exists today, and has existed for a long period of time. So it may not be a completely meaningful result in the sense that you couldn’t have operated a business all that time ago.

Nonetheless, it’s not meaningless either. It’s a result that says that price returns auto-correlate in assets, albeit weekly, and they have done for a long period of time. This is consistent with a behavioral bias that exists, that humans and investors have a propensity to follow trends.

[00:19:13] Alan

Yeah, fair enough. Niels?

[00:19:16] Niels

I want us to go back a little bit here, Phil, because I find it very interesting that your roots really are trend following and then you add other things to your business, for a period of time, and then you come back and you double down on trend and launch a dedicated vehicle for that. So a couple of questions. First of all, I would also like to say that I really like the fact that you’ve done all this research and it confirms your belief in trend.

So one question I had is that when you look at your own business today, does it surprise you that trend is a relatively small piece of it or is that just the way it is because trend is generally a smaller piece than that? The second question would be, when you then relaunched a trend strategy in ’13, was it really just taking what you already had or did you come at trend with a slightly different approach than where you came from originally?

[00:20:21] Philip

Okay, great question. Why is it a smaller piece? I think that trend following has struggled in adopting a role in traditional portfolios for a number of reasons. It’s not as big a part of institutional portfolios as I would say that it should be. It’s a strategy, as I’ve said, that ticks all of the boxes. It’s genuinely diversifying, genuinely uncorrelated, and even has features which are good in the tails. But I think that people do struggle with it too. If you’re not a disciplined investor and you are a performance chaser, then because of this profile, this positively skewed profile of the P&L, that’s a performance chaser’s nightmare.

I think a lot of people, they invest in it after one of these accelerations, so they are performance chasing, perhaps, after the global financial crisis, after 2014, or even after 2022. They stick around for a few years and remain “patient” before the next acceleration, but by the time the next acceleration comes about they’ve redeemed. So when they look back on their P&L, through trend following, it’s actually quite poor. I think that people struggle with that. Whereas my view is that it should really be a buy and hold. Obviously, I’m biased.

In terms of if I was managing my own money (I’ve had this question before) “Out of all the strategies that you run, if you could only have one strategy, which would it be?” I think it’s a no brainer. It would be trend following because it ticks the most boxes.

Obviously, our business has evolved massively beyond trend following, but it’s not because we don’t like it. It is because we feel that we need to try and stabilize and increase the level of return so we’re diversifying into different approaches, different datasets, different techniques, etc., etc. So, it’s a huge research process. Given the choice, out of everything that we do, I think trend following is probably the one that does tick the most boxes.

Your question about how we implemented it, did we do anything different? I think our added value in our dedicated trend following fund is (well, there are many things). firstly, not falling into the pitfalls of in-sample overfitting. There are many, many ways of doing trend following. You can google trend following and there’ll be tons of different ways of doing it. It’s actually quite easy to overfit, generate more performance in-sample and then you don’t get performance out of sample. So, that is something that requires a lot of experience to exploit trend following in the right way such that you get that out-of-sample format, which is what we’re all trying to achieve.

Then I think, as well, that our added value is everything that goes into building a product, which is to say, portfolio construction, execution, all of the research that we do in those areas applies in our trend following product as well. So that is how we set it up, is to try and use what we feel are good implementation skills to build a better trend follower.

[00:25:00] Niels

We may come back to that a little later in terms of how you may think of yourself as being slightly different. But I wanted to bring up something that we brought up with everyone, which is this Cliff Asness paper that came out last year, where the key takeaways, really, is that on one hand, Cliff is saying that maybe we as an industry, we’re just becoming too concerned about Sharpe. We’re doing things to trend essentially maybe more for business reasons rather than for actual performance reasons and that may be not a great thing.

He also talked about whether we, as an industry or as trend followers, have a dual mandate. One is to generate absolute return, but also to generate this crisis alpha (I know you already mentioned that). So, I would love to dig into one of these topics and see where we go with that.

[00:25:58] Philip

Sure. We have a lot to say about convexity. We’ve written papers on the subject and it’s an interesting subject that trend following offers. In our last paper on the subject, we had these rather beautiful payoff functions that you could generate that looked like straddle payoffs. So you could make trend following look like option payoffs. But I think what people don’t realize is that convexity is really a mechanical effect in trend following. So even if you apply a trend to a random walk, sometimes the random walk wanders away too far and if that wandering persists on a timescale of your trend, then your trend following approach is guaranteed to make money.

So that’s where your large infrequent gains come from. The rest of the time, because it’s a random walk, then your overall P&L is zero. So the rest of the time, you’re losing. You get small, frequent losses, but large infrequent gains. It’s starting to look like an option payoff. That’s really the difference with applying it to a financial time series.

You get this positive P&L, when you apply it to a diversified pool of instruments, and that’s because of this behavioral bias, as I said. So, it’s kind of a minor miracle really that you get to a positive Sharpe ratio plus this convexity if you contrast that with options where premium is very pricey. You’ll get a very strongly negative P&L if you buy options, but of course, the convexity is different in its nature. It’s only convexity with respect to very long and drawn-out protracted moves in the underlying.

Our experience with clients; we had some clients that would come to us to talk about the Sharpe ratio of trend following, we would have some clients come to talk to us about the convexity. Based on our understanding of convexity, we saw that as a business opportunity to try to, in a very transparent fashion, maximize convexity in a product for a small reduction in Sharpe.

So we say to our clients, we’re going to reduce your Sharpe ratio and use that discounted chart to try and cram in as much convexity as possible. Convexity gets reduced by trend followers by doing a few things. I think the most interesting one is the diversification, firstly, that CTAs do trend following on a diversified pool of instruments. Now, your convexity on the S&P (which is what everybody is looking for) is limited to trending on the S&P. If you diversified beyond trending on the S&P, then you’re not adding extra convexity.

Let’s imagine that you add orange juice to your S&P, then you are adding convexity on orange juice (which maybe some people are interested in). Mostly people are interested in downside equity risk. So adding trend following on orange juice does not help you in terms of your convexity.

Then we had, also, some clients that struggled with this 2018 scenario where you get equities rallying, and then equities crashing (like the VIX-mageddon event), and as a trend follower, your P&L then looks very correlated with equities. We also added a constraint to the portfolio construction to make sure that we have no exposure to equities. You end up with a product which is anti-correlated with equities. All of this was in production through COVID and we had very good results through COVID.

Coming back to Cliff’s paper (as you can tell, I’m on first-name terms with Cliff), AQR, they’re a great firm. They’re very good at timing their marketing documents in particular. I would say that our view is that if you want trend following as a hedge, then we have a convexity-enhanced product. If you want trend following for performance, then we have a different product.

If you want something with a better Sharpe ratio and it’s less correlated to trend following, then we have our flagship offering. So it all depends on what the client wants. I don’t see why managed futures should be restricted just to trend following. We have many, many strategies in our flagship that are doing many things that have nothing to do with trend following at all. I don’t know, as an esoteric example like using weather data to forecast commodities that really has nothing to do with trend.

[00:31:34] Alan

Diving into research, you did make a statement about the pitfalls of in-sample overfitting, which I think I get what you’re alluding to, but it may be useful for listeners to explain exactly what that might be in a trend following context. Is it just literally doing the backtest and selecting the parameters that historically worked best? What is a better way of doing it or what’s the right way to do research in terms of a trend portfolio?

[00:32:11] Philip

I would say that in the business that I’m in, in-sample bias is the biggest problem. It’s the most pernicious bias that we have in doing research. You kind of see when you have younger researchers that turn up, that are very enthusiastic, they’ve put together P&Ls. After doing this for 23 years, the amount of P&Ls that you see that are going up, and yet in production, in realized performance, were nowhere near that. So, there’s clearly something going on. You don’t see what you see in your backtests so you always take backtests with a big pinch of salt.

I think the way around this, there are many things that we do to try and get around this problem; trying as much as we can to do out-of-sample tests, to try to limit the size of the parameter set to get an understanding of the model. Let’s say trend following is a good example because if you understand where the origin of the effect comes from (so it’s a behavioral bias, it makes a lot of sense), then that gives you extra confidence that it’s an effect that is going to persist in the future. So all of that helps to generate out-of-sample performance.

In the context of the trend, what I mean really is that you can fool yourself. Let’s say you have your benchmark strategy and you’re trying to improve on your benchmark strategy. So you start changing parameters, maybe there’s a Monday effect; that of trend following works better on a Monday, so you over-allocate on a Monday and under-allocate on the other days. But then you start to realize that it’s not a statistically significant effect.

But the problem is that you can sort of have many small or close statistically significant effects and then you start accumulating them and your P&L gets better and better and better in the past and that’s not reflected in the future. So it’s all of these types of problems and challenges that we have to deal with.

[00:34:53] Alan

Taking that mindset to the non-trend strategies, do you need an intuition around why they work or not in terms of convincing yourselves of their validity?

[00:35:07] Philip

It depends on the strategy. We do very different types of research. Often, it’s as a function of time scale of the strategy. On the shorter-term side in our flagship program, we have shorter-term strategies. There, you can do things with higher levels of statistical significance that you’re less reliant on actually having an understanding. For machine learning strategies as well, when it starts getting more black boxy, you would like to understand what the black box is doing. But if the performance is so good, it kind of negates the need to get an understanding of it. But then as you go out slower and slower strategies, you don’t really have that luxury. So there, you really need to get an understanding of what you’re doing; an understanding of what’s driving the effect. So yeah, it really depends from model to model.

[00:36:25] Alan

Going back to trend, obviously, you want to avoid overfitting, tinkering with the parameters. So what are the avenues for enhancing a trend program via research given that you want to do as little as possible with it by the sense of it?

[00:36:45] Philip

To some extent, yeah, the research that we do is very much dominated by work on our flagship program. Never say never in terms of what we do with the trend. We have made some changes in trend following. We have a component which is more like a cross asset trend within the trend program, but it’s small in comparison to what I would call single assets or traditional trend following approach. So really, you’re right that the development, per se, in terms of the research in signal development, is nowhere near what we do elsewhere in the firm. I would say the development is more in the other areas like execution and like portfolio construction.

[00:37:48] Alan

Fair enough. You touched on machine learning there. Some of our guests have found that helpful, some less so. I suppose what you can get with machine learning is that adaptive nature, shifting to a new environment. Is that not something you’ve found success with applying that in trend following in terms of picking up that markets are trending faster or slower? Anything around that?

[00:38:23] Philip

Not particularly on timescales, but you can identify regimes where trend following works more or less well, but it’s not a strong effect. It’s nowhere near as strong as the actual trend following effect itself. So we have toyed with that. I think the biggest effect is trend following, the second biggest effect is “buy the dip” type strategies. The problem with buy the dip, especially when you get to a crisis (so the S&P is tanking), and then you go in and you buy the dip. Where is the bottom?

You can quickly find yourself in a position where you’ve come out of the trend, and perhaps you’re even going long the S&P, and then the S&P is crashing. That’s a very awkward position to be in. So these types of strategies, they’re not trivial. We have found value in machine learning. It’s not been a revolution. It’s been a slow burn. I think that when we talk to clients, they think that machine learning is something quite new. When I was doing my PhD, I was using machine learning. It’s been around for a long time.

[00:39:46] Alan

Sure. Yeah. No, absolutely.

[00:39:48] Philip

I think what has changed is really the availability of technology, computing power, the availability of recruits that know a lot about this, that know how to use the technology, the fact that you can do things on perhaps two lines of code. When I first started, we had to code a lot of stuff ourselves and now there’s a lot of libraries that you can use to do this type of thing. But yeah, we see value in machine learning less so on the strategy side, but a lot in execution, for example, where you’re looking for shorter term predictors. Machine learning is very useful there. It is definitely a growing component of our research.

[00:40:49] Niels

I wanted to ask you, first, a very selfish question, because it’s a question I get a lot actually. So I’d love to hear your thoughts on this. If we just take 2022 as an example, everybody talks about it as being a great year for trend following, but it was also a year where we saw a lot of dispersion in returns between “trend followers”.

What do you think when you look at the landscape of peers? What’s the main difference you think you have or trend followers have between them? Is it just the markets we trade? Is it times of speed of trend? What are the things that you think really make a difference between one trend follower to another?

[00:41:46] Philip

That’s a good question. So yes, timescale is one. I would say the portfolio of instruments; these are key differentiators and we can delve more deeply into that. Actually, in preparation for this discussion today, we looked at the dispersion of CTAs, historically, and tried to see whether or not there was a pattern there. There is a slight effect, it seems, that when volatility goes up, dispersion goes up, which I found a little surprising, but it’s not that big an effect. So, I think what’s happening there is that the timescales may be slightly different.

When you’re in a crisis, some people get into a position to benefit from that earlier than others. There is a difference in the timescales. Our trend following approach is on a mid to long-term frequency. The reason that we steer clear of the shorter timeframes is one of cost. So you actually see a very stable performance, in terms of Sharpe ratio, across the timescales of the trend as long as you don’t go too short. You do see a big difference when you incorporate costs, because on the shorter timeframe, you get more turnover and that incurs more cost.

But on our product where we’re trying to create equity downside protection, we did fill in that gap, in terms of timescales, to try to get more protection on a range of timescales. So, I think it depends what you’re trying to achieve as to what timescale you operate on, but there’s clearly a difference across CTAs. In terms of the instruments that people trade, there’s been a move towards what’s now become known as alternative markets. Perhaps you could also refer to them as being illiquid markets. It’s a very interesting development, I think.

If you consider alternative markets and what the optimal amount of instruments that you should be trading is, then you should have as many markets as possible if they’re decorrelated. If they’re uncorrelated instruments, then you should just add everything.

If you consider the common correlation among markets, then the power of diversification is curtailed beyond a certain number of instruments. But then liquidity is a real problem. I think that our focus has been on cost and liquidity perhaps more than others.

When you add illiquid instruments, that’s quite a risky endeavor. The point that I would make is that when you’re trading illiquid instruments, what we’ve seen is that the performance of alternative market CTAs has been quite good as there’s a flow of AUMs that can move into that space. Potentially, the performance is actually coming from a flow of AUM (a steady flow of AUMs) that’s pushing prices in the direction of their position. So, you can generate performance like that. If ever that flow of AUMs were to stop, then there could be quite a brutal decay of the impact in illiquid markets that will send performance in the opposite direction. So it’s an interesting development.

I would say, as well, that the danger is that the more diversified your pool of instruments becomes, the less convexity you get, the less defensiveness you get, the less equity downside protection you’re going to get. With the example that I gave of orange juice, trending on orange juice does not help you to protect against the equity S&P downside. As you go to more and more esoteric portfolios, then you get less and less convexity, and that’s perhaps what we saw last year in 2022. The more illiquid alternative market offerings, their performance albeit positive was not as positive as on the liquid markets.

[00:46:56] Niels

Yeah, absolutely. Now, one thing I’ve certainly learned to appreciate in my career, in this industry, is that the devil, really, is in the detail. Yet most of the conversations, I think most of the understanding or the way people think about trend following, often goes to the actual kind of signal generation. Are you using moving averages, breakout, time series momentum? What are you doing? But there are many other parts of what we do.

I’d love to hear your thoughts about the importance you put on things like correlation, volatility, the choices you make about these things. How much do you think that actually moves the needle relative to whether you’re using one type of trend following or another, so to speak? I’d love to hear your thoughts about that.

[00:47:48] Philip

Yes. I think I would group that under a category of risk control. How do you do your risk control? So yeah, historically, we’ve put a lot of effort into understanding correlations, into cleaning correlation matrices, reducing the amount of noise in the measurement of correlations, and that certainly helps. We see value, I think, in that you need to be able to estimate the correlations and clean your correlation matrices, but you also need to get estimates of volatility. I think the volatility estimates need to be somewhat more prompt in terms of being able to deliver your targeted risk.

So what we are trying to do with our risk management is that we’re trying to keep risk stable, we’re trying to hit a target, and we don’t do that for fun. It does give us better risk-adjusted returns. When you backtest, keeping risk stable is better for outcomes. If you think about what we’re trying to maximize with our risk management, I think what we’re trying to maximize is Sharpe while controlling the tails. You could think of it as being perhaps Sharpe over kurtosis, like fat tails, a measure of fat tails. So, you’re trying to minimize the fat tails and maximize the Sharpe ratio.

We do that algorithmically. I think that would come as no surprise. Because our risk modeling is backward-looking, necessarily, there are things that are happening in the market that we’re aware of that our risk model is not aware of. So we do, also, try to forecast gaps in the market, the potential for tail events, and we do that by sitting around a table in a committee.

So I sit on that committee. We’ve discussed such things as like the Bank of Japan recently coming out of the yield curve control that they’ve been operating, also through COVID, what could potentially happen in terms of tail events, and looking forward to the debt ceiling that is coming up in the US, could that potentially create a tail event. So yeah, obviously, we’re doing everything algorithmically.

But I think that that side, trying to understand and being aware of what’s happening in the market is also very important.

[00:50:48] Niels

Okay. Yeah, that was actually one question I wanted to just maybe ask you because I wasn’t entirely sure. So what you say is you’re aware of these, but do you actually let the systems run as they should or do you intervene?

[00:51:04] Philip

We intervene.

[00:51:05] Niels

You intervene?

[00:51:06] Philip

Yes. So for example, the Bank of Japan, what they were doing in terms of yield curve control, we stopped trading the JGB.

[00:51:17] Alan

I just wanted to move on and talk about the role of trend and managed futures in a multi-asset portfolio because you has touched on a few of these topics already. Obviously, you mentioned, I don’t know if you call it vanilla trend, your flagship trend and then you have the more defensive trend with the capped equity beta. Do you see that as driven by investor demand? If you were running a simple multi-asset portfolio of stocks, bonds and adding something to it, do you think it’s better to go with something where the trend has the equity exposure curtailed or not?

[00:52:09] Philip

So we did, obviously, a lot of research on this point. We have a dedicated trend following program, we have a convexity-enhanced program (whatever you want to call it), we have equities, and you could even throw in a 60/40 which tends to be quite correlated with equities anyway (but whatever your preferred benchmark is), and then try and see, based on various metrics, what the results are when you combine everything together. So that required a certain amount of thought as to what metrics it is that you’re really trying to maximize.

The way that we thought about the problem, rather than looking at Sharpe ratio, is to look at Sharpe ratio also incorporating drawdowns in various forms because it’s not easy to get a handle on drawdowns. You can look up the Sterling ratio and that is like a gain over drawdowns rather than a gain over volatility, but it’s similar in spirit to what we were trying to do. But then how do you define drawdowns? That’s really the problem. We had various definitions of drawdowns.

In the absence of having a dogmatic view, then you average over several different definitions and that’s what we did. Yes, it confirmed that our equity downside protector was better, in the mix ,when picking these types of metrics that don’t incorporate drawdowns. So that’s the way that we dealt with it.

[00:54:22] Alan

So if you were looking at the total portfolio, say 60/40 plus trend, if you were just looking at the compound annual growth rate (forgetting about drawdown), you would be better off with your normal trend unadjusted. Is that fair enough?

[00:54:41] Philip

Better off or not much different.

[00:54:44] Alan

Okay. Yeah. You talked a bit about the convexity, in this example, of adding orange juice or not. Our experience is often that actually, when you get these equity drawdowns or equity sell-offs, it’s actually the non-equity markets where you get the big move. So COVID, for example, the commodity markets started showing evidence of the problem. Obviously, they were very China-sensitive in January, February of that year, and obviously, you tend to get moves in bonds. So is that not an important source of the return profile which gives managed futures and trend following this convex profile?

[00:55:29] Philip

That is indeed the case. Strictly speaking, the convexity comes from trending on equities, but what you know is that things will be correlated, so you get correlations in the tails in particular. Yeah, I think through COVID, the bigger contributor was actually fixed income. So, yes, you get convexity through correlations with other instruments, but the more esoteric the instrument gets, the less the correlation is and the less effect you get.

These moves in commodities, we wrote a paper on the fact that… I think what actually happens with commodities, if you’re trending on commodities, you actually get a hedge against inflation, so convexity against inflation. If you try and model inflation, then you see that it’s very well-modeled by commodities. It makes a lot of sense. So if you just regress CPI with all of your instruments, you’ll see that commodities are really dominating the explanation of CPI.

So when you’re trending on commodities, you can think of it as being trending on inflation. In the big moves in inflation up and down, then you get protection. So you’re getting that convexity on inflation as well, which is also a great thing I think probably most people don’t realize.

[00:57:04] Alan

Yeah. Maybe just one final one on this topic of using trend within a multi-asset portfolio. Obviously, we’ve talked about how you capped the equity beta, and really, what we’re talking about here is dealing with 2017, 2018 type scenario where trend following becomes correlated with equities and then you’re worried about both of them turning down at the same point.

Have you looked at other things, other ways of possibly solving this problem of either adjusting the risk (so maybe having less risk in trend at that time), or also maybe adjusting the speed (still trading other markets but making some other adjustment to how trend operates) to avoid that concurrent drawdown with trend and equities?

[00:57:55] Philip

We have looked through the history of the firm at such things, but yeah, it’s very, very difficult to time and to know ahead of time. It’s tautological, as you look back, that one timescale was better than others, but to actually do that causally (or what I call causally) without the benefit of hindsight, it’s very difficult to capture that effect.

[00:58:28] Niels

Yeah. Before we start wrapping up our conversation, Phil, just a couple of things. One of the themes that we are picking up from our conversations is really that a lot of the larger managers as well, a lot of the managers in the CTA index with whom we’re speaking to, saw flows head out of trend last year, most likely because investors were rebalancing their portfolios, etc. But it kind of also touches on a little bit the point about just generally capacity among managers and strategies, etc. So I’d love to hear your thoughts a little bit about what’s a good measure.

How do managers find out whether they are near or at capacity in general? The other thing I wanted to ask you, because maybe some of the assets flew out of some of the larger managers, but we did see that replication products did see an inflow probably from a completely different investor base, namely, the individual investors. But since you are much more of a quant than Alan and I can claim, I wanted to ask you, what are the pitfalls, potential pitfalls, of doing replication on managers when you only look at some kind of regression of returns, when you’re not actually building your own model beneath that?

I guess my own concern is that we may see a big inflow into these replicator products because they promise that they can deliver everything but the fees, so to speak, which we can always debate. But it would be a great shame.

I like the people who do it. They come on the podcast on a regular basis. So I have to be politically correct when I say this. But of course, one of their pitches is that you have this manager risk, right? Just because XYZ did well last year, you don’t know if that person is going to do well next year. Sure.

But what about the replicator risk? There must be a risk somewhere if you’re choosing one way of doing replication that could also get you into trouble. So that was a lot of questions in one sentence, Phil, so pick whatever you remember from it and let us know.

[01:00:53] Philip

I’m going back to how you started this question, so rebalancing. Indeed, we have seen some rebalancing. That’s definitely the case. I find it interesting that clients will rebalance their equity portfolio, so when they see downside risk in equities (they get losses from equities) and then they reinvest. You never see that with trend following. When trend following underperforms, they never reinvest in trend following. I think, again, it’s kind of a sign that there’s not really traction in institutional portfolios for trend following.

On the subject of capacity, the capacity of trend following, it’s an interesting question. It’s always an interesting question where your capacity is for a strategy or for a product or program. As a scientist, you’d kind of like to run an experiment and go way past your capacity to see where your capacity is. But obviously, that’s not a good business model. But trend following, it’s a funny animal really because it’s not an arbitrage. You’re not taking advantage of a mispricing. Just modeling trend following, you do see that if you add more trend followers to a market, then you get better performance.

The mechanism for removing the trend signal from a market through an excessive amount of people is not really there. So the capacity, I would say, for mid to long-term trends is higher than most strategies, because firstly, it’s a directional strategy so you don’t really have to run market neutral portfolios. It’s something that naturally lends itself to generating a lot of risk, so you don’t need a lot of growth to be able to do it. Also the fact that you’re trading the most liquid instruments in the world. These are the deepest markets in the world so the capacity of trend as a strategy is actually high.

When we estimate capacity, I think that what we tend to do is really just base it on our experiences as much as anything. We also, as a firm, have the benefit of running higher frequency, faster strategies. So we have a huge database of the trades that we’ve done, often quite sizable trades, so we feel like we have a control over our execution. So, potentially kind of scaling up the strategy in your backtest, you feel more comfortable doing that.

On the last point, the replication issue, we have not built a replicator. I’ll just say that, again, what we’re doing is we feel that we can offer the market a very good trend following, dedicated trend following program. The risk of these replicators, I would say, is really the implementation skills required to do this. It’s a non-negligible amount of work that gets put in, many millions of dollars invested into building these types of programs.

So I would say that the main risk, in investing in a replicator, is that you’re not investing with someone who has that skill set. Yes, you can regress and they can offer you a product with very low fees, but just be careful of the fact that the cost may well be in the fact that you do not deliver the performance longer term. That’s what I would say.

[01:05:38] Niels

I appreciate that. Final question, Phil. That is just something we wanted to ask everyone in this series, and that is, what’s the one thing you hear about trend following that you disagree with the most?

[01:05:57] Philip

I’ll give you perhaps a few.

[01:06:00] Niels

Oh, yeah, sure. Probably more than one.

[01:06:04] Philip

Firstly, that it’s easy. I really don’t think that it’s easy. You could come up with a naive P&L, but then going from that through to something which is implemented and actually generating a reasonable degree of risk-adjusted returns, that’s not easy. Perhaps I’m repeating myself, but don’t underestimate the importance of implementation skills.

Also, I think that there’s perhaps a level of naivety on the convexity. The downside protection that you get from the trend is that it may be a mechanical effect, but it still remains statistical. It’s not the same thing as options.

If you want the insurance, then you have to buy options and pay the premium and that gives you guarantee to payout when the thing that you’re insuring against actually happens, meaning the equity crash. In the case of the trend, you get that protection to path-dependent things when markets are in that long and protracted drawdown which is kind of often the case. You see crises, crises tend to play out on long timescales. You look at the global financial crisis, that’s what happened and that’s why the trend did very well. The CTA business did less well through COVID and I think that it was a much faster move.

With our convexity product, we’ve managed to capture some of that, but because it was a short-lived move or even like February 2018 as well, that’s really too quick to be caught by most to CTAs. So you just have to be careful of that. There is protection, but it’s a certain type of protection.

[01:08:28] Niels

Yeah, absolutely. Phil, this was a great way to wrap up a fascinating conversation. Thanks so much for being on the podcast and for sharing your thoughts and insights with us and we hope we can do this again sometime in the future. To all of you listening today, I hope you were able to take something from today’s conversation onto your own investment journey. If you did, please share these episodes with your friends and colleagues.

From Alan and me, thanks so much for listening. We look forward to being back with you on the next episode of Top Traders Unplugged as we continue our deep dive into the CTA industry. In the meantime, go check out the show notes for this episode and all the other resources that you can find on our website. As usual, not least, take care of yourself and take care of each other.

[01:09:13] 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 on 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.