266 Systematic Investor Series ft. Rob Carver – October 21st, 2023

[00:00:00] Intro

You’re about to join Niels Kaastrup-Larsen on a raw and honest journey into the world of systematic investing and learn about the most dependable and consistent, yet often overlooked investment strategy. Welcome to the Systematic Investor series.

[00:00:23] Alan

Welcome back to Top Traders Unplugged, where each week, we take the pulse of the markets from the perspective of a rules-based investor. It’s Alan Dunne here again sitting in for Niels for the final time this week. Niels is back next week. This week, I’m delighted to be joined by Rob. Rob, how are you? Good to see you.

[00:00:39] Rob

I’m well thank you. I think it’s been quite a while since I’ve been on, probably since maybe June. Back then, it was sunny and warm and nice, and now it’s cold, wet and miserable. So that’s the way it goes I guess with the English weather.

[00:00:53] Alan

Winter is creeping in, but you do have the prospect of England in the Rugby World Cup to warm the heart and give you something to look forward to the weekend, I guess.

[00:01:01] Rob

Well, that’s right. By the time the listeners listen to this, they’ll probably already know the result. But yeah, England are lucky in that we’re the only team from the so-called home nations to have stayed into this stage. We did have quite a lucky win last weekend to be fair, and I think we’re going to have to do pretty well to beat our opposition this weekend, but fingers crossed anyway. Fingers crossed.

[00:01:24] Alan

England playing the world champion, South Africa, so yeah, that’s really formidable.

[00:01:27] Rob

Yeah. It’s not going to be an easy match. All credit to Ireland last weekend. They played really well and came really close to it to go through against really top-class opposition. So yeah, we’ll see how it goes.

[00:01:38] Alan

I was in Paris, so it was fantastic. We did a whole podcast to dissect the reasons for Ireland’s repeated exits at quarterfinal stage. But no, Ireland played really well and New Zealand were fantastic too.

[00:01:54] Rob

All the quarterfinal matches were really good actually, really close.

[00:01:57] Alan

Yeah, really looking forward to that later this weekend as well. So yeah, it’s been a while since you’ve been on. What’s been on your radar since the summer?

[00:02:06] Rob

Yeah, it’s been an interesting few months P&L wise. I was just kind of pulling up my account curve as I usually do before I come on. So I think when I last spoke to you, I was probably down for the year actually. I think like a lot of people in the trend following and CTA space generally, I had this shock in March when everything moved against me on the back in Silicon Valley. Then things started to gradually creep up.

I think last time I was on, I probably made some of that money back but was still down for the year. But I have to say, things have gone really well actually. Since mid-July, performance has been pretty good, went on a real tear up to mid-September, then a bit of a dip and now it’s coming back again. So actually, I’m quite pleased with how things have gone. I’m up about 10% actually for the year, which is pretty reasonable and I think compares well with industry benchmarks as well.

But I think it’s perhaps more interesting to kind of dig in a little bit and just look at where some of those gains and losses are coming from. So on a sector basis, actually, the big news (and maybe we’ll come to that in a minute), but the big news in the market obviously is the bond market, fixed income market. As an ex fixed income trader myself, I’ve always got a close interest in what’s going on there. Yeah, so my second-best performing sector actually was the bond and interest rate sector. That’s contributed significantly positively to P&L.

One of the best-performing individual markets was US 3-year bonds for example. Interestingly, the best sector of all though wasn’t actually fixed income but something that hasn’t been in the news quite as much, and that’s FX. So I can see that I made quite a lot of money out of the renminbi/dollar, the yen/dollar, and also the Aussie/dollar. So they’re kind of Asian. Asian currency complex versus the dollar has actually been quite profitable for me. So it’s kind of interesting because that hadn’t really been on my radar.

That’s why it’s interesting to look at these reports and just see, oh, I didn’t even know I was trading that market potentially. But then on the other end of the line, the equity markets have been quite frustrating this year. There’s not really been any clear trends, been almost a classic kind of sawtooth. Not a bad year if you’re buying and holding. A big part of my overall investment portfolio is buying and holding and not trading. But for someone trying to catch trends, it’s been quite frustrating. So that’s been not great.

Another report that I run is to take a step back from my overall portfolio and just look at the market generally. What I do is I look at all of the markets I track, which now it’s about 250 futures markets. So that’s pretty much all the liquid ones plus quite a lot of ones that aren’t so liquid actually. Then what I do is I look at the price movements and I adjust them by volatility, so I get a kind of normalized view, which is perhaps more interesting than just outright up or down, which you probably see crypto at either end of the spectrum depending on what’s been going on there.

So the best-performing market this year, actually, interestingly, on both an outright basis and on a vol-adjusted basis is orange juice. Orange juice has absolutely flown. Unfortunately, I have not had a position this year, so I’ve not benefited from that directly at least. Interestingly, all the other markets that are right up there at the top of that ranking, they’re all agricultural markets. So they are cocoa there, I’ve got robusta coffee, I’ve got both kinds of sugar up there. So it’s a lot of markets that I think, listening to the news and stuff, there’s inevitably a lot of focus and concentration on the financial markets.

Most people can tell you what the S&P has done this year or what the bond markets and done. It’s being shoved down our throats the whole time. But perhaps certainly on the more obscure corners of Twitter are people talking about orange juice or sugar or what have you. In terms of markets that have done quite badly, the Israeli shekel has done quite badly, and that’s not really a huge surprise given what’s in the news at the moment.

This isn’t a podcast about current affairs, but obviously, what’s going on in the Middle East is pretty terrible. As someone who lived in the Middle East in my formative years as a child, it’s pretty horrific. But this isn’t the time or the place to discuss the rights or the wrongs, so let’s leave that there.

Also doing quite badly, at least in terms of outright performance, are the volatility markets. So what that means in practice is that volatility has actually being crushed this year because being short vol, being short the VIX futures essentially means you’re betting on the market getting safer. So interestingly, actually, that short vol position has done very well.

Now, it’s not necessarily that the actual level of vol has got massively crushed, because a lot of the returns for being short, the VIX futures comes from effectively the difference between what people predict volatility will be and what volatility turns out. That’s called the vol premium, and that tends to be significantly biased. In other words, people tend to overestimate what vol is going to be.

That means when you’re holding the futures in a short position, you’ll tend to benefit from a rollover or carry effect. So I would guess, I’ve not done the math, but I would guess that maybe half that return is because the absolute level of vol has fallen over the year, which may seem surprising given I guess we’ve been in some kind of fairly global type war situation because we’ve had Ukraine going on.

But obviously, the recent events obviously you’d expect to have led to more volatility markets. That doesn’t seem to have really happened. So that’s also all fallen off quite a lot. And then interestingly, another market that’s done poorly that most people won’t even have on their radar is coal. Interestingly, it’s actually Newcastle coal. I’m actually going to Newcastle for a few days this weekend for some holiday. So a bit of a coincidence there.

[00:08:03] Alan

Interesting. A chance to get the local perspective on the coal market.

[00:08:08] Rob

Exactly. Yeah. One more thing very quickly that I like to do is just look at my overall market positioning, which to me is sort of interesting. So I’m still substantially short the bond market. As you would expect, that’s my biggest single position. I’m short US 20 years, I’m short bunds. But I do have some other positions in there. I’ve still got some of those currency positions on. I’m short Swissy, dollar-yen, and also pound, I think. Is that a short position or a long position? It’s a short position.

So I’m not being very patriotic. Although I am supporting England in the rugby, I’m actually betting against the pound at the moment, so that’s interesting. So yeah, it’s been an interesting year. It’s not over yet, but it’s been a positive one but not a knockout of the park one with quite a lot of surprises and apart from, obviously, the short in the bond market, no real massive standout trend like we had with commodity prices say 18 months ago.

[00:09:13] Alan

When we look at the industry, just to recap quickly on industry performance, month to date, slightly negative. SocGen CTA down 90 basis points, SocGen Trend more or less flat, down 3 basis points, and Short-Term Traders up 22 basis points. Year to date, SocGen CTA up 88 basis points, SocGen Trend of 92 basis points, and SocGen Short-Term Traders down 1.72%. So, as an allocator, when you speak to a manager and you hear they are up 10 in a year where the industry is flat, you’ll ask, what are you doing differently? What’s your perspective as to why this environment is favoring your program?

[00:09:52] Rob

Wow. The first thing to say is obviously I don’t accept outside money. So sorry, I can’t take that call from the allocators. Although, I might let them buy me dinner before breaking the bad news to them.

[00:10:05] Alan

The response you would normally say was I don’t know what everybody else is doing, but you probably do have a pretty good idea of what everybody is doing.

[00:10:10] Rob

Wow, so I guess there’s a number of different answers to that question. The most honest answer is performance on a year-to-year basis is luck over short periods of time. I’ve tracked my performance since I began trading my own money which, and it’s terrifying say this, but is about nine and a half years now, actually, since I started trading my own money in this little futures portfolio. I stopped trading institutional money.

I do pretty well versus the indices, probably a bit ahead of them. But that doesn’t mean on a year-to-year basis, there could be huge difference. There’s been years I’ve been much worse than the index and years I’ve done much better. The other thing is I’m probably less of a pure trend follower than a lot of people. So I have quite a lot of other signals in my portfolio. What that will mean is that in a year when trend just blows the doors off, I will tend to lag.

In a year when trend isn’t doing so badly than all the other stuff in my portfolio, I’m still probably 60% trend, but that other 40% will help push everything up so it smoothens out a bit.

[00:11:10] Alan

If you look at your P&L attribution, I don’t know if you have it, is that what you are seeing if you look at it by different trading strategies? Is non-trend doing well this year?

[00:11:21] Rob

That’s a report I haven’t run recently. That would be my expectation, yeah. Because as I said, if you look across the markets, there’s nowhere really, apart from bonds, where there’s been a really clear trend. And it’s quite hard to make money shorting bonds because the yield curve is still upward sloping enough. The carry is against you. You don’t make as much money shorting bonds on the way down as you do being long bonds on the way out.

Apart from that, equities, there isn’t really anything there. A few commodities have done really well, but it’s not like the whole commodity complex. It’s been very selective. So it’s not really surprising to me that it’s been a difficult year for trend. I would be surprised if the trend following part of my portfolio was doing amazingly well. My very strong intuition is that it’s the other stuff that’s helping.

[00:12:16] Alan

Maybe a bit more of a fundamental question, but you did mention being a former fixed income trader. There is an ongoing debate I guess around the term premium and what it should be. It was negative, now it’s positive. I just briefly opened up Robert Armstrong Unhedged on the FT there. Some people are saying, well, there shouldn’t be a term premium anyway. You always get these debates where the term premium is knocking around.

What’s your perspective? If you were trying to model a yield curve, would you have a baseline assumption for what the term premium would be? How do you think about that?

[00:12:53] Rob

I guess the thing about any kind of risk premia is it’s very easy to say that (so I disagree that there should be no term premium), but it’s common sense there should be term premium because (putting on my economist hat for a second) people have a preference for liquidity. So, they like to have money now. If they’re going to lose money and have it back in the future, to me, the further in the future that is, they’re probably going to want to have a higher reward for that risk.

Now, of course, there are technical reasons why, for example, when you get quite a long way out on the yield curve, particularly in the UK, there are technical reasons why the yield curve then tends to become flat or even inverted to do with basically pension funds hedging long-term maturity risk. I guess intuitively, you’re like, well, if I’m going to lend money out for 10 years, it doesn’t really matter if it’s 10 years or 20 years, it’s still a long way in the future.

But to me, if you’re lending your money out for 2 years or 10 years, there should be a term premium there. Similarly, it’s clear that there should be a premium to holding risky assets like equities, for example. The hard thing becomes in saying what level those things should be at. For example, if you go back to, I don’t know, about the early 2000s certainly, and certainly before that, maybe the late ’90s, equity returns had been incredible, and the equity premium therefore had been very high.

When economists tried to explain the equity premia in terms of people’s aversion to risk, they just couldn’t do it. The level of equity premium was just too big. Now, the good news for them is that since then, we had a series of market crashes and the equity premium is not looking as good, and perhaps, certainly on a backward-looking basis, perhaps now at a level where it makes more sense. So it’s really hard.

The best job you can do is just look at a lot of data and say, well, on average over the last 30, 40 years, it looks like this thing should be at roughly this level. That’s the approach taken by, it’s not quite the same thing, but something like the Shiller CAPE ratio. You have a kind of opinion on where the long-term level should be. You just take a really long average and say, well, on average, it looks like (let’s go back to bonds) ten-year bonds seemed to return, on average, about 2.5% more than 2-year bonds on average, on an outright basis.

You do that exercise and then you can then say, well, if that’s the case, then the carry of the premia is very low or very high, and therefore, I expect that to revert or increase to that, and that’s part of my valuation model. As we know, these things can go a long way from those levels very, very, very quickly. Now, one thing that is quite interesting actually, I did a blog post recently. I’m not going to discuss in great detail because it’s not really a trading blog post. But what I was doing was looking at the risk-adjusted returns across and within asset classes, which is a kind of interesting empirical question. It makes sense if you’re an investor but not if you’re a trader, so not necessarily that relevant.

What I found really interesting was that in most asset classes, there isn’t really a strong relationship between risk and return. In other words, generally speaking, you don’t really get rewarded for more risk within asset classes. You do across asset classes. So you tend to make more money in say commodities which are riskier than say bonds or FX. Another way of saying that is, if you look at the Sharpe ratio of different asset classes, it’s broadly the same. So you broadly get rewarded for return and risk across asset class. Within asset classes, not so much.

A lot of people did a lot of work in equities on this and I thought I’d have a look at it across asset classes in different asset classes of futures. What I found was, yeah, generally speaking, there isn’t really much of a relationship there, and the only exception is bonds. In bonds, there really is a very strong relationship between risk and return. In other words, generally speaking, longer-duration bonds which are riskier because they’ve got high duration, duration is just sensitive to interest rates, so they’re always going to be riskier generally speaking.

Higher-duration bonds earn more than lower-duration bonds. On a risk-adjusted basis, if you look at the Sharpe ratios, they’re fairly similar. That’s something that’s always unique about the asset class. So it does look like in bonds, the evidence for there being some kind of term premium over long periods of time is fairly solidly there, I would say. So, yeah. I’ve not really answered your question.

[00:17:41] Alan

It’s interesting. I would have guessed that maybe the return for the volatility in bonds is less. In a sense, if you look at something like the really long duration treasuries or something like TLT and the size of the drawdown we’ve seen there, which is what? I guess, 40%, 50% or something, maybe more. The volatility there is certainly as high as equities, I think, if not higher. I’m not sure what it is over really long periods. But I would have thought that the returns are obviously not as high as equities. But then bonds are a kind of an unusual asset in the sense that the more they sell off, the future returns look more attractive.

[00:18:27] Rob

That’s supposedly true of equities as well. It’s just that the math isn’t as obvious, right?

[00:18:30] Alan

Yes, exactly.

[00:18:31] Rob

So in reality, it’s not that bonds are an interesting kind of asset. It’s more that bond and equity investors think differently. That’s the way I think.

[00:18:40] Alan

I guess so. Yeah.

[00:18:41] Rob

Just to go back to what you said, what’s interesting is I’m looking at this graph more closely, and actually, that relationship I described does seem to break down beyond about 10, 20 years. As I said right at the start of the discussion, weird things do seem to start happening at the very long end of the yield curve. So it’s quite a strong relationship. If I look at the returns for say US 2s, 5s, 10s, and 20s, there’s pretty much a straight line there. You do seem to get rewarded pretty consistently for the additional risk you’re taking on.

But then if I look at 30 years, actually, they do worse than 5 years, so I don’t know. There’s an awful lot of caveats to this kind of analysis. One is that the lengths of history are different for each of these instruments because the futures haven’t been around as long, so I’d need to dig in more and see if there is an effect there. But generally speaking, I would say that you do tend to get (when I’ve looked at this in more detail, and this is a while ago now, this was like 10 years ago, so there’s more data in the dataset), generally speaking, it does seem to be that shorter duration bonds on a Sharpe ratio basis do a little bit better than longer duration bonds. As I’ve just said, very long duration bonds do seem to be a not a very good way of accessing interest rate risk because you just get a lot of volatility and not much returns.

[00:20:03] Alan

I guess that might make sense in terms of, obviously, with the long-duration bonds, historically, you’ve had that positive convexity, so you have that most property that there might be that desire for people to hold them in their portfolios, which would justify a lower Sharpe than the lower duration bonds. Would that be reasonable?

[00:20:23] Rob

Yeah. One thing I found interesting was why two, three years ago, it was only like, I don’t know, the Austrian government was issuing 100-year bonds. That was the trade of the millennium, right? The UK Government, say for example compared to something outside the US, do issue longer duration bonds, and therefore, their funding cost is kind of locked in for longer. But while we were just pushing out 30, 40, 50, 100-year bonds, people would have been buying this paper. The people stupid enough to buy them would now be sitting on 70%, 80%, 90% loss, but the government would have locked in cheap funding for five or six generations. But anyway, that’s another story.

[00:21:14] Alan

Good stuff. Well, we’ve got a lot of interesting topics. You reached out to your fan base on Twitter and they came back with quite a number of interesting topics, so we may as well just go into them one by one in terms of the questions.

The first one was from Andrej. He’s asking, “What’s your take on systematic strategies complementing trends such as buy fear, sell greed short-term mean reversion type strategies to fill the potholes in the equity curve?” He says, “Alan and Nick talked about this last week.” So yeah, it’s the general point about, and I guess, as you say, you’ve got them in your own portfolio, what’s your take on reversion-type strategies as a diversifier to trend following?

[00:21:59] Rob

Well, the nice thing about reversion-type strategies is that at the same timeframe, they’re obviously perfectly negatively correlated with trend following. So they would provide you the best possible diversification. Now, of course, if you were to just say, let’s say you trend follow with a holding period of say three months roughly, if you were to just flip that signal around and do a mean reversion trade, you would be perfectly negatively correlated, but you’d lose money, at least in the backtest, because trend makes money, right?

So what you’ve got to be looking for is other places where trend doesn’t work so well and where mean reversion works so well. So you’ve got to move away from, certainly, the time frequency is one thing you can do. The other thing you can do is go into more relative value trading. So it might be that you’ve got two assets that trend well. But actually, there’s a relationship between them, a relative value trade between them which is mean reverting.

Actually, you see that lot in the fixed income space. People trade, for example, butterflies, which would be something like say the 2, 5 and 10-year point on the yield curve. If you plot the relationship between those and you vol adjust it or beta adjust it or duration adjust it, it is quite mean reverting. So that’s potentially mean reverting trading in the relative value space. But obviously, what the question is pointing at is the idea of that it’s sort of staying with a kind of outright system, so not doing relative value, but moving into a different time space. So you can either go slower or faster.

We talked about the idea of mean reversion to some equilibrium. That’s your classic kind of slow value trade. It’s quite easy to implement something like this in a CTA-type portfolio. You just do something like take the price for the last five years. You probably want to normalize it in some way so that the series has, to be technical for a second, got stationary properties you can transform it into expected volatility standardized series. If I’m allowed to plug just for a second, this is described in my book, my latest book.

Basically, you do that and you look at say the last five years and you say, okay, if the price in this adjusted way is below the price the last five years, we buy, otherwise, we sell. This is exactly the kind of behavior when talking about these bond guys say, “Oh, look, the price has gone down a lot, therefore, we buy. The price has gone up a lot, therefore, we sell.” You’re basically assuming that nothing about the market has changed to make that kind of equilibrium price any different. You’re just interested in buying the dips and selling the highs.

Now, the downside with that is that because it’s a slow trading strategy, generally speaking, there’s almost a mathematical law that the slower you trade, the lower your expected returns will be because you’ve just got fewer opportunities. You’re not visiting the table in the casino quite as many times. You’re talking about something that rather than say across a diversified futures portfolio, I might expect to hit a Sharpe of say 1 trend following. This is more like 0.2, and it’s pretty close to being statistically insignificant so it’s not really a strategy you can trade. There’s very strong evidence for this being a great thing to do, it’s something you basically you have to just trust it’s going to work, and you never put a big amount of money into it for all of those reasons.

The other thing you can do is push to the shorter end of the spectrum. Really, now we’re getting to where the question is going. Now, it does look like, if you look at the performance of trend following strategies at different holding periods, there’s kind of a sweet spot around a holding period of maybe a month say, and it depends on the asset class. So in equities, it’s a lot slower. Go slower than that and the performance starts to come down a little bit, but not dramatically. So even with holding periods of a year, it’s still pretty good.

Go faster than that (ignoring costs as well, by the way, this is before costs, because obviously, the faster you go, the more costs you have to pay. I’ll come on to that in a second), before costs, the performance starts to degrade. When you’re down to holding periods of a week or maybe a couple of weeks or a few days, around that region, performance doesn’t look very good at all.

Now, poor trend following performance doesn’t necessarily translate into a profitable mean reversion strategy even though they’re the opposite sides of the coin. One reason is costs, because obviously, these things are going to be trading in a much higher frequency than something that’s got a one-month holding period. Now, that’s maybe not the end of the world because it might be that you just have to cut down the number of things you’re trading to small the universe.

So you may end up with a lot fewer commodity markets in there and a lot more equity markets, because the equity indices, on average, tend to have cheaper trading costs. That’s one way of getting around that problem. But that automatically is a less attractive idea because one of the great wonderful things about trend following is its ability to multiply performance massively. Individual instruments don’t do that well. A portfolio of 100 instruments does really well. With these mean reversion portfolios, you’re maybe not going to get that if you can’t trade everything.

Now, the other thing is you’re not just looking for a weak or slightly negative trend following performance because, ignoring costs, that’s just going to translate into mean reversion profits of zero or slightly positive. It’s not necessarily that attractive. If it involves trend following, it doesn’t just need to decay to nothing as you go faster, it needs to actually go negative and negative enough to pay for trading costs and to become an attractive proposition.

I’ve done a lot of research on this, but I have to say more in a theoretical sense. I’ve never quite got around to yet implementing a superfast trend following strategy myself. So what we can do is say, well, we know there’s something that definitely works as a mean reversion strategy, and that’s high frequency trading. Because you’re basically hoping to buy at the bid, sell at the offer and just do that 100 times a second, all day, every day.

But obviously, the barriers to entry to that, the technology required and so on, it’s something that’s not just out of scope for me as someone trading my own money, but actually out of scope really for a lot of institutional CTAs because they’re not necessarily going to have the right people or the right skills, the right technology. If they do want to get into that business, they’re going to have to make a huge investment. You think about the people they’re competing with, people like Citadel, for example, it’s quite a high moat those guys have got around them.

There are startup HFTs that do well with limited amounts of capital. But again, that’s the key thing. It’s limited amounts of capital. If you’re a billion-dollar CTA and you can trade 10 million of that as HFT capital, is it really worth it? It’s not going to move the needle, even if in its isolation it’s a highly profitable strategy. So we’re looking to go a bit slower than that.

So I did some analysis really. I found for example that if you look at something called the autocorrelation of returns, so this is just a fancy way of saying given what the return was one period ago, what is the period correlation between that and the correlation now? If returns are positively autocorrelated, that means, for example, if the price went up last month, then is more likely to go up this month. Well, I’ve just described a trend following strategy basically. So that’s the sort of thing we’re looking for.

The autocorrelation of futures returns for one to two hours is quite strongly negative. So that’s indicative that mean reversion will probably work well at that frequency. Then there’s effectively it kind of gets a bit worse and worse, and then it’s meandering around.

There are times when it’s more weakly negative, so again, something like three or four days it looks okay, and then as we know, when we get up to two weeks, one month, but trend following is working, so the autocorrelation has gone positive. So it looks on paper like stuff that for me is very short-term. Obviously, as a high-frequency trader, an hour is like an eon. But for me, that just is insane. It seems insanely quick. It looks like mean reversion would work very well there.

I have done some work showing mean reversion works pretty well for the five, six days as well, but it’s much harder than trend following. So we’ve discussed costs, that’s one reason why it’s much harder. There are things you can do with your execution strategy to try and reduce your costs without the more difficult but from a kind of operational technological point of view, it’s much harder.

The other thing that may not be so obvious is actually risk management mean reversion strategy is also much harder. Because the nice things about trend following strategies is they kind of do their own risk management. So you get into a trend and then there’s two main ways people tend to play it. One is to put a stop loss in. But I say, you know, I don’t need a stop loss. I just wait for the trend to reverse itself and close my position and get into the new trend on the other side.

You can’t really do that with mean reversion because if you get into a mean reversion trade and it then moves in your favor, well, you’re going to close the position at some point because you’ve reached the equilibrium. If it moves against you, your mean reversion formula says, oh, buy more, this is great, let’s have more of this, let’s just fill our boots. That’s just obviously not necessarily a great thing to do because if the market keeps moving, you’re going to get murdered. So you probably do need an explicit stop loss in that situation.

Then you get into this slightly tricky situation where you’ve got this extra parameter to calibrate and the thing is, it’s well worth it to say, “Oh, I’m going to close a trade now,” but then when do you reopen it? You’ve got to wait because it looks like it’s a great trade now. So do you wait for the price to move one cycle back in and out? It gets really, really complicated and much harder to backtest and not over-parameterize.

There are ways around that and there are things I’ve looked at. These are things like trend following filters as an alternative. Stop losses don’t work pretty well. So it’s kind of something that I like the idea of, but for me personally, I know it’s going to be quite a lot of work. I’m not sure that the extra return is worth it.

Very quickly, one thing that’s much more promising though is using these kinds of very short-term signals as some kind of overlay to your execution. So in very simple terms, the way that might work is this. Suppose that you think that prices mean revert on say a daily basis and you’ve got a daily trading system. Something you could do that’s really simple is open up the other market in the morning.

Let’s say your daily system wants to sell. Well, you’ll only sell if the price on the last day has changed in accordance to the mean reversion system. So you’re never actually trading the mean reversion system, the short-term system outright, which would be too expensive, but you use it as an overlay to your trend following system. So that’s another way to use it.

[00:33:12] Alan

Yeah. As you say, intuitively, it is appealing. Things like overbought, oversold, that would have been the classic technical perspective. Trade it with a trend and then looking for signs of trend exhaustion and looking at momentum indicators and things like that. Presumably, you haven’t found anything in that domain that has worked.

[00:33:30] Rob

That’s a whole part of the trading world that I kind of stay away from to be honest because a lot of that stuff’s very, very subjective. I tend to follow most of these indicators fall into one of three camps. They’re either basically doing the same as something I already use but in a slightly more woolly way, or they just don’t work, or they’re subjective. So yeah, I generally stay away from that stuff. That doesn’t mean to say it doesn’t work. It’s just not my expertise.

[00:34:07] Alan

Obviously, as you say, with respect to the longer-term reversion, you do see that a bit in terms of even more fundamentally oriented. You touched on currencies and the yen. Obviously, a lot of people selling it, now the yen is very undervalued. So if you had a multi-system, multi-strat, presumably, you’d still be short yen and the trend basis but then I guess you would start to balance out with more long yen positions. But as you say, then how do you risk manage that? The further it goes, the more value there is there. So I guess there’s probably a sizing element presuming you would never run these on an equal risk allocation basis.

[00:34:45] Rob

No way. I don’t have the short-term stop. I do actually have a long-term kind of value time indicator in my system, but it’s got quite a small risk weighting. So yes, at the moment, it’s probably buying a little bit of yen. But it’s overwhelmed with the size of the position from my trend system and probably my other systems as well. So it’s just moving the needle a little bit because yeah, it’s such a weak signal and also hard to risk manage it. It seems the worst possible thing you can trade. I believe it deserves a small allocation because of the negative correlation, but it doesn’t make any sense to make it 50-50 as you say.

[00:35:21] Alan

Yeah. Okay, fair enough. We’ll move on to the next question from Harry. It’s around medium and high-frequency trading strategies. He says, “Medium and high-frequency trading strategies can presumably achieve much higher Sharpe ratios than lower frequency strategies employed by CTAs. So can you speak about the lure of high-frequency strategies for traders with modest capital? How much research have you done?” So you’ve talked a little bit about that, but maybe just focusing particularly on I guess all of the different types of strategies, including trend, on short-term time horizons.

[00:35:54] Rob

Yeah, Harry is right. I did say already that the more times you get to visit the casino, the more money you’re going to make as a rule of thumb. Actually, there’s something they call the law of active management that says basically, if you half your holding period and all other things being equal, your returns will increase by the square root of 2. So that means if you go from say an average holding period of a month, which is about four weeks, down to a week, then in theory, your returns could double.

Now, there’s a lot of work being done by the words in theory there. So obviously, costs, we’ve talked about it already, how that will have an effect. The other thing is of course you’re assuming that you can find the things that are as good as your current spectrum. So for example, if you were to just speed up your trend following system, certainly, as I said, I see trend doing worse as I speed it up from one month down to three weeks, two weeks, one week.

Some people have shown that, for example, any sort of intraday trend strategies seem to do well. So maybe there’s something there. But the problem is, again, costs, and costs actually for very short trend following strategies are really, really bad because the kind of execution tactics you can use to try and reduce your costs, actually, you can’t use them. Because basically, let’s say you’re trading trend with a holding period of something silly like, I don’t know, an hour. To me, that seems silly.

My current execution strategy is basically patient; is happy to kind of just chill out and spend the whole day executing an order if necessary, because you know what? I’ve got a one month holding period. I know that one day is not going to make any difference to my alpha. But on the other hand, I could get the trade more cheaply if I just chill out. But if you only hold the position for an hour, then you can’t do that. You’ve got maybe a minute, three, four, maybe five minutes tops, to get into that position before the alpha starts to decay.

That means you’ve got to be a lot more aggressive in your execution. You’ve got to cross the spread, and that means you’re going to be paying those spreads every single time you trade. Your trading costs just go through the roof. So I think it’s very hard to trade very fast trend following strategies, except, again, perhaps as an overlay on top of a slower system as an execution overlay. That may work.

Again, it’s something that seems great, but actually, when you break it down, I guess, part of it (if I’m being honest), part of it is path dependence, right? If I was starting from scratch with no kind of previous experience, maybe I would be like, actually, I’ll build a fast trend following system or a fast trading system full stop. But because I look at my current system and it’s pretty good, it works a particular way.

I kind of feel I understand it. I kind of feel I’m exploiting well-learned effects and I know that my trading costs are never going to be excessively high, and I’ve got potentially a huge investment of time and effort and then risk of capital to do something that’s very, very different but I don’t feel I’ve got any expertise in. So that’s the truth. So any research I do in this area really is going to be about potentially putting in some kind of execution overlay.

[00:39:11] Alan

But is that because you come from a large institutional background and that’s your whole experience?

[00:39:15] Rob

Yeah. As I said, I think that’s probably quite a lot to do with it. The allure of the pot of gold at the end of the short-term higher, not high, but higher frequency trading strategy, I have been trying to find it. Occasionally, I kind of get my spade out and I dig around, I do some backtests. Occasionally, I find something that looks promising. But then it’s not promising enough because, for example, I could only trade a limited number of markets. So automatically, I lose all this diversification benefit I’ve got.

The problem I don’t have as I’ve got quite a small account, by institutional standards, quite large by retail standards, but quite small by institutional standards, so capacity is not an issue for me. So that wouldn’t be an issue for me. That’s a good thing. The effort required for what I feel would be quite a modest improvement. The best case scenario is I’d find something I could trade with three or four equity markets, hive off part of my capital to trade just those markets.

Yes, that would probably be profitable, but the opportunity cost of losing that capital into a less diversified, even though it’s a higher performing strategy and a risky one in a sense because of the higher costs. Yeah. So I’m not saying don’t do it, but things that are profitable are not easy. Otherwise, we’d all be rich. Put that on my tombstone.

[00:40:50] Alan

Yeah. There is a sense that with certain strategies, being capacity-constrained can obviously work at lower levels of AUM.

[00:40:56] Rob

I’m not sure if that’s 100% true.

[00:40:59] Alan

But you’re saying that maybe fast trend following isn’t maybe necessarily one of those, or it might be but it’s just the operational headache. You’d actually have to sit there every day and monitor your portfolio probably a lot more than you currently do.

[00:41:13] Rob

People do fast trend following, or at least people market themselves as fast trend followers, and some of them make money. So there’s clearly a business there. But it’s like anything else. A lot of the time, I get asked on the podcast, “Rob, why don’t you trade equities? Jerry trade equities.” I’m like, well, yeah, if I had a team of three or four people working for me, I probably would trade equities. But as an individual, the additional benefit I get from trading those equities is not sufficient for the work that would be put in.

That doesn’t mean by the way that I think that trend following individual equities is wrong or bad or a silly thing. I don’t think any of those things is true. I’m a bit more skeptical about fast trend following than I am about slow trend following of individual equities, so the bar is even higher in a sense, because I’m not 100% convinced there’s something there. Whereas I know for a fact, because I’ve done it in the past in an institutional environment, that if I was to use my existing system on individual equities, it would be profitable. It’s just that the work involved is quite high.

[00:42:16] Alan

So what would you say is the advantage? If you’re trading say a million-dollar account versus a billion-dollar account, what are the things you can do at the million or maybe 10 million? Is it having a larger risk position in some of the smaller markets? Is that the big advantage would you say for the smaller account?

[00:42:35] Rob

I would say yeah, that’s definitely one of the big advantages. So when I was trading at AHL, it was a 30 billion rather than 1 billion account. There’s a sweet spot, I would say, somewhere between 50, 100, and probably 250 million, I would say, because my problem is I can’t trade everything I want to trade because of account sizing, because of position sizing. So you go down to an extreme level where you’ve just got $1,000 account, you can probably only trade a couple of instruments that have a low enough margin requirement to trade.

As you get more money, things improve because you can add more and more instruments. That’s great. Somewhere around 50 million, you’re probably able to take a position, at least a one-contract or two-contract position in pretty much every liquid futures market in the world. That’s great. You’re unconstrained in terms of your allocation, so you can allocate just as much to cheese as you can to S&P 500. As you go from 50 to 250, that equation doesn’t change that much.

So you maybe can’t put quite as much in cheese as you can in S&P 500 when you get to 250, but it’s not really enough of a difference to make material difference. Then on the journey between 250 and 30 billion, you can no longer have an unconstrained approach capacity-wise to risk allocation. So when you get to 30 billion, you’re going to have a lot more in S&P 500 than you are in cheese. That isn’t because you think S&P 500 is a better market than cheese, it’s purely a capacity issue. So $1 billion is probably a bit north of that optimum point, but not too far off there.

So yeah, as an allocator, there are advantages to being in a $30 billion rather than a $250 million dollar fund as well. Because again, for example, you might not have the heft, for example, to trade OTC markets where you need isness and things like that. As a $250 million fund, you’d be able to get in the door of the prime broker, but you wouldn’t actually be able to employ the people required to kind of deal with the operational side of running an OTC trading business. So there are still advantages as you go above the 250 mark, but the main disadvantage is absolutely you can hit capacity constraints as you try and diversify across market space.

[00:45:08] Alan

Makes sense. Okay, we move on to the next question from Ravi. “Can you please talk about techniques to prevent overfitting of a strategy? Is it possible to quantify fitting of a strategy to data? More importantly, how can we detect overfitting? And then separately, how can we measure the robustness of a strategy?”

[00:45:27] Rob

So what’s interesting about Ravi’s question is actually, he’s put scare quotes around every other word. So he said fitting, he put fitting in scare quotes. He’s got protect in scare quotes. He’s also put robustness in scare quotes. I’m wondering whether this is because he doesn’t believe that these things exist or whether he’s not sure of the terminology. But anyway, that’s fine. This is something that we could talk for hours, days, years on. At some point, I will probably write another book on this subject because it’s something I get asked about quite a lot.

So the way I think about overfitting is basically what you’re doing is you’re assuming that the future is going to be too much like the past. So we have to assume the future is going to be a bit like the past. Because we say, well, for example, we backtest trend following strategies and we find in the past they do well. So we kind of think, okay, well in the future, trend following will do well, so we’ll carry on doing it. There’s a big assumption there, clearly, but I’d say all systematic traders have to make that assumption. That’s fine.

What isn’t fine is saying something like, well, if I look at gold, I find that using a 32 and 87-day crossover with a stop loss of 1.6 times the annual volatility with a stop profit of 7.8 times the price, and also, I’ll close my position if the third Fibonacci retracement on a Wednesday, it’s a negative value, and the moon is full. That’s overfitting, because what you’ve basically done is looked at your backtest and said, well, when that incredible confluence of things happened, gold did really well in the backtest, and it was probably on the 23rd of July in 1984 or something like that. You’re like, that was the best day to have gold in the backtest on all these things happened.

That’s overfitting, clearly, because you’re assuming that the future is going to be exactly like it was in the past, and that’s clearly not going to happen. Now, the line between those two extremes of the continuum is obviously very wide. It’s not easy to say exactly where the point is where you’re going wrong. It’s easy to say theoretically where you’re going wrong. Basically, the more fitting you do, the more you improve your in-sample performance, the backtest performance of a strategy, the worst your strategy will do out of sample.

What you want to do, what you’re aiming to do is maximize the out-of-sample performance of your strategy because that’s where you make money. No one pays you for a backtest unless you publish it in an academic journal perhaps or unless you’re working for a fund that foolishly has some kind of backtest related pay structure. So you want to maximize your out-of-sample performance.

That means you want to basically do just enough fitting so you’re capturing, for example, we were talking about speed of trend, so you want to be, for example, doing just enough fitting so you’re not trading the very fastest signals where the trend doesn’t do so well and is also very expensive and you’re probably not trading the very slowest signals where performance starts to degrade. You want to be kind of in the middle somewhere. That’s fine. That’s good. That would be a good amount of fitting to do. You don’t want to be going beyond that point.

Now, to actually detect it, there are a number of different ways you can detect it. One way is to look at the performance of your strategy across different sub-samples of the past. You can either literally do that by saying how did it do in the 1980s, how did it do in the 1970s? Or you can pull random years out or random days or months out with a technique which is known as bootstrapping. What you basically do is look at the distribution performance of your strategy across the different, let’s call them, alternative histories.

What you’re looking for is something that does pretty well across lots of different alternative histories. If, on the other hand, your performance is very peak, in other words, it does really well for the alternative histories that were very much like the history you’ve actually had, then you’ve probably overfit and you should be very wary of running that. Because the chances are that you’re sitting on the peak of that high mountain very low, you’re more likely on the other sides of the deep, despondent valleys of overfitting, which is actually a good title for the book I might write on the subject.

That’s also effectively a measure of robustness because what you’re saying is, is that the strategy robust to conditions in the market changing and not being exactly like they were in the past? You can take things further. So for example, you know, a lot of people, interestingly, 10, 12 years ago, a lot of people were saying, “Wwell, we don’t think that CTAs are going to do very well when interest rates start rising, when that starts happening.”

It took many years, many years for that to happen, the reason being that a lot of performance has come from being short bonds and it’s hard to make money. Sorry, being long bonds, it’s hard to make money being short bonds, we talked about that. So one thing I actually did as an exercise was to look at the conditional distribution of returns of the strategies, conditional on what interest rates were doing, when they are low, when they are high, when they are rising, when they are falling.

Basically, if our strategies had not been robust, we’d have seen amazing performance when interest rates were falling. As it was, the distribution of returns between those different environments was actually fairly stable. It was a bit better when interest rates were falling, but it wasn’t like terrible in the other states. That hopefully gave people confidence that when interest rates start rising, CTAs could potentially still carry on doing well. You can judge for yourself whether that was an accurate statement.

[00:51:18] Alan

Well, we’ve already one version of the history of…

[00:51:20] Rob

Yeah, exactly. So that’s a very, very brief discussion of what is, as I said, a massive topic. There are all kinds of fancy techniques of doing it. But the main thing is to, if you find yourself running 100 backtests and perfecting your strategy, then you’re almost certainly overfitting. If you’ve just done one or two, then you’re probably okay. That’s a good rule of thumb.

[00:51:45] Alan

Okay. You mentioned bootstrapping there, which is kind of a technique of, as you say, randomly plucking the data from the past. Obviously, it’s statistically valid. Some people might say, well, the returns occurred in a particular order in the past given the sequence of events in the world.

[00:52:02] Rob

That’s why you can use a technique called block bootstrapping, where you basically sample blocks of time. So you might sample years, for example, rather than just days. You’re actually right because what you want to do is preserve the order of the returns within those periods. So obviously, if you’re trend following with a monthly time horizon, and you start bootstrapping days or even months, then you’re going to lose that autocorrelation and just going to get noise back. So you then just do it with years.

The downside of that of course is there’s only a limited number of years in your dataset. So you get less information from what you’re doing. But it’s still better than just running a single backtest.

[00:52:02] Alan

Fair enough. We’ve got a couple of questions from Tommy, so we will go onto those given we’re moving along on time. So first one, “What’s your take on the AHL trend ETF?” It’s called AHLT. We touched on this a little bit a couple of weeks ago with Andrew. Yeah, quickly, have you had a look at that? Any thoughts on AHL moving into the ETF space?

[00:53:09] Rob

So obviously, I have a lot of loyalty to AHL. As I often say, I worked for them for quite a while and they paid for my house effectively. So I’m still a big fan of them, although I don’t own any shares or anything like that, and so my interest is purely sentimental rather than financial. I’m also a big fan of people democratizing, let’s call it, the alternative asset space, let’s call it that. So I’m a big fan of that. I’m a big fan of ETFs as well. I’m quite jealous as a UK, a European-based investor of the variety of ETFs in the US in particular. It seems to be a lot easier to branch out. So I like the principle of the idea.

Now, the problem with the ETF implementation is that it often comes with some restrictions. You just can’t be as unconstrained as you are in a standard managed account structure for example, or in an offshore fund. Now, I don’t know because I’m not an expert on the legal side of things whether some of the choices that have been made on the design of this product are due to those restrictions or whether they were deliberate design decisions. So I’ll say that upfront.

With all that in mind, the expense ratio, I think it’s 95 basis points, which is okay. For an alternative beta fund, it seems pretty reasonable. Obviously, that’s probably cheaper than investing in a classic hedge fund structure. I should also declare an interest in that one of the portfolio managers is an old colleague and a very old friend of mine, so I’m obviously biased there as well.

It doesn’t seem to have many instruments inside it. Looking at the material I’m looking at, it has 20. To me, that feels low. As I said, one of the big advantages for the trend following universe is diversification. Twenty to me seems on the low side. So that’s my main bugbear with this product. As I said, that may be a legal regulatory compliance issue as to why that’s there.

Otherwise, yeah, I’m a big fan of the minimum account size to either trade your own portfolio or even to invest in a say diversified fund of funds is quite high. I’m a big fan of something where someone can put $500 into something that’s going to give them exposure to something they wouldn’t otherwise have exposure to.

[00:55:39] Alan

If you were to guess, putting you on the spot here, what would be the Sharpe degradation going from this all bells and whistles, diversified managed futures programmed trading 200 markets with maybe trend and non-trend down to a 20-market trend program?

[00:56:00] Rob

I don’t need to guess because I’ve done the analysis. So in expectation, it’s probably in order of magnitude of 2. So you’d be cutting like a 1.5 Sharpe to a 0.75 Sharpe, something like that.

[00:56:12] Alan

Okay. But you still think it could be as high as 0.75 for the trend?

[00:56:17] Rob

Well, I mean 1 to 0.5. That’s probably more realistic, actually. Yeah, and of course, that’s an expectation, right? So you could get lucky and your 20 markets would be doing brilliantly. But for example, this year, if one of those 20 markets was orange juice, you’d look amazing. You’d probably be doing better that 100..

If you didn’t have orange juice and you were playing against someone who did have orange juice, obviously, in expectation, you’d be doing worse. So on a year-to-year basis, there’s a lot of luck involved as to whether you do worse or better with a bigger portfolio. But in expectation, I would say you’re going to cut your performance by, let’s be generous, between a third and a half, I would say.

[00:57:02] Alan

Obviously, we know with something like that, presumably, launching an ETF, the idea is to have big assets, so it’s always going to have more risk in fixed income and the liquid markets. So given that you know that in advance, does that tilt the number for even a little bit more?

[00:57:21] Rob

Well, not necessarily. These alternative asset ETFs, obviously, it’d be amazing if it became like SPY and was, whatever, half a trillion dollars or whatever ridiculous number it is. These alternative ETFs tend not to be that big. So, obviously, it would be great for AHL as a business if it did become very big. That would be superb. AHL, let’s say, I don’t know the exact numbers because I’m not following the business that closely, but say it’s a $30 billion fund, they would do well to have a $1 billion ETF in alternative assets. That’s increasing their market footprint by 3%.

So if we look at it holistically, it’s not really making a lot of difference. This is an interesting kind of question that we used to discuss, because actually, let’s say that you’ve got a $30 billion ETF and you’re constrained in your allocation to say cheese, you launch a new small fund, do you allow it to be completely unconstrained and say, yeah, you can have relatively a lot more cheese, probably not much cheese in the grand scheme of things still small fund. Yeah, I like cheese. That’s why it’s my favorite example of a market.

Do you say, well, actually, if we look at our portfolio holistically across all of our funds, then we really ought to have the same quite small allocation to cheese because in our $30 billion-plus epsilon fund, we don’t want our market footprint in any market to be too big, therefore, we should have a small allocation to cheese. Essentially, otherwise, what you’re allowing is the people who are coming into smaller fund, you’re giving them preferential treatment effectively over your other investors, for example. So that’s not necessarily a kind of theoretical question. That’s actually potentially a business question or a legal question.

[00:59:21] Alan

Yeah. Okay, final one just to wrap up. We’re up on time. From Tommy again, “Trend seems to be the perfect diversifier until it’s not. In which scenarios do you expect equities and trend to be highly positively correlated.”

[00:59:34] Rob

When equities are going up.

[00:59:38] Alan

When CTAs are long equities.

[00:59:39] Rob

When CTAs are long equities and where also there are other assets which have got strong trends and therefore providing. So it might be that. Let’s make it really simple. Let’s just make it super simple. Suppose you’re a CTA and you trade just two markets, bonds and equities, and you ask yourself that question. Well, clearly, it’s going to be the case when equities are going up and bonds are going up strongly and you’ve got a long position in both or when equities are going up and bonds are going down, you’ve got a strong long and a strong short position. It’s going to be one of those two scenarios.

It’s not going to be the case when equities are going down and bonds are going up or bonds are going down. It’s not going to be the case when both markets are going sideways. This kind of comes back to what I was saying earlier about performance during the year in equities not being great. You might look at the year-to-date performance of equities and say equities have done pretty well, CTAs on average haven’t done that well.

But actually, it doesn’t mean that equities have trended necessarily either. So it’s not just that equities need to have gone up, they need to have gone up with a clear trend. So yeah, if Tommy is asking for some kind of macroeconomic analysis, then my friend has come to the wrong place. I’m just going to give you the naive, stupid answer, which is when they’re both going up.

[01:01:00] Alan

Yeah. We’ve seen that obviously in the data. Obviously, trend and equities tend to be positively correlated in an equity bull market and tend to be negatively correlated in an equity bear market in broad terms, which is exactly what you’re saying. So, very good. Well, we’re up on time. We’ve managed to get through all of the questions bang on time, so that was an achievement in itself. So thanks very much, Rob. Great to chat to you.

From all of us here on Top Traders Unplugged, thanks very much. Niels is back next week, so it’ll be business as usual. Take care and speak to you soon.

[01:01:35] Ending

Thanks for listening to the Systematic Investor podcast series. If you enjoyed this series, go on over to iTunes and leave an honest rating and review and be sure to listen to all the other episodes from Top Traders Unplugged. If you have questions about systematic investing, send us an email with the word “question” in the subject line to info@toptradersunplugged.com, and we’ll try to get it on the show.

Remember, all the discussion that we 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.

Thanks for spending some of your valuable time with us and we’ll see you on the next episode of the Systematic Investor.