The Surprising Factors Behind CTA Performance: Is Less More? ft. Rob Carver

[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] Niels

Welcome and welcome back to this week’s edition of the Systematic Investor series. With Rob Carver and I, Niels Kaastrup-Larsen, where each week we take the pulse of the global market through the lens of a rules based investor.

Rob, it is wonderful to have you back on this autumn Thursday. Hope you’re doing well. How are things in the UK?

[00:00:41] Rob

The sun is shining, it’s quite warm, quite pleasant. So yeah, it’s a very nice autumn day here and it does seem like a long time since we’ve spoken, although it’s probably not been that long. I’m looking forward to getting through some questions and interesting topics today.

[00:00:57] Niels

Yeah, it’s a pretty full agenda I have to say. Before we get into all of that, And there are quite a few questions, we’ve got a couple of really important topics as usual. It’s always good to start with just sort of what’s been on our minds recently, what’s been on our radar, what are the kind of things that caught our attention. So, I’m going to let you go first with something that you found interesting.

[00:01:22] Rob

Yeah, yeah, so, I don’t know if you remember, I’ve got a couple of pet hates actually. So, one of my pet hates is people blaming CTAs for everything. And so, I was very disappointed to see an article on the Financial Times website this week, which was blaming CTAs for the volatility in oil prices.

So, I actually wrote something for the FT, a while ago now, sort of going do some maths and explaining like why, you know, I think it was Goldman’s that had blamed CTAs for basically moving the equity markets and with the maths was simply saying, look, we’re just not big enough.

I mean we’d love it, we’d love it, wouldn’t we, if CTA’s had sort of $5 trillion under management and could genuinely be responsible for moving the markets in such a big way. But sadly, that’s not the case. We’re actually quite small players, and even with a bit of leverage, which obviously we use, we’re not really big enough to move things in markets.

There might be a few pockets of markets out there where they’re not especially liquid and where there’s not much kind of real commercial activity going on and there’s a lot of sort of speculative activity. But most CTAs are quite sensible and don’t sort of try and trade in size. That means they would move markets.

So, I was disappointed but not really surprised to see yet another kind of, you know, oil price volatility is down, all down to CTAs. I think it’s very unlikely.

[00:02:47] Niels

So, was it just a sort of a general comment in the article that we, as CTAs, are behind, or was there anything in terms of, oh, they were pushing the price in a certain direction? Because I was looking at it when you sent over the link, and first of all, if they’re talking about recently, I mean oil prices, they’re up only about 1% in the last month. Right?

Now, since April, it’s been selling off. I think we hit a high of $87, down to $65. We’re at $72 right now. And actually, I suspect that oil hasn’t been a very good market for CTAs this year.

I mean, a lot of models would have tried to go long early on, been disappointed by trying that, and now we’re probably kind of flat, slightly short. But not even all oil contracts, is what I’m trying to say, are the same this year. It’s kind of diverse. So. Yeah…

[00:03:45] Rob

It does make me laugh because, I mean, I’m with sort of giving a hint to people that later on in this episode we’re going to speak about one of my other bugbears, which is replication of CTAs. You know, if you do any kind of basic replication of position sizing and behavior, and you’ve got an understanding of how things like momentum and volatility drive our positioning, you know, you do that and you say, well, actually it’s quite unlikely that we’ve got very big positions on at all.

One of the things I sometimes do is run down my positions. I’ve not done that, not today, but I can very quickly tell you that if I do look at my positioning as of now, where am I? Do I have even a position in crude? I do have a position in crude, but it’s very small and I’ve got a small short on. And that’s what you’d expect, I think probably, given the price activity.

You can sort of plausibly say that given a really long, smooth trend you’d expect CTAs to buy into with big positions because the trend is smooth. Therefore, volatility is low. Therefore, leverage would be high. In a position where that trend is sharply reversed, you potentially expect a lot of CTAs to rush the door and sort of increase the size of any move downwards. But that’s clearly not what’s happened in oil.

So yeah, there we go. We’re the bogeyman behind every door. But you open the door and there’s just a small child dressed in a bed sheet. There you go. There’s a Halloween analogy for you.

[00:05:19] Niels

Absolutely. And before we jump to another article that you just mentioned before we hit record, I just want to mention one thing to people because I do think it’s such an informative example. And that’s that this year in Cocoa, where a lot of people thought, well, the price has gone up by you know, X 100%. Clearly that must be CTAs pushing the price higher.

Well, in fact we’ve sold about 90% of our cocoa into that rally because, as you mentioned, when vol expands we contract our position size. So, sometimes these moves are actually completely opposite of what we are being, you know, blamed for.

Anyways, you mentioned another article that I’m not aware of or familiar with.

[00:05:59] Rob

Yeah, this is even more disappointing. Yeah, this is my day to be disappointed, I’m afraid. So, this came out in risk.net, which is the online version of Risk magazine. Risk magazine is a very kind of prestigious magazine and famous for publishing articles full of lots of equations. So, it’s very much at the kind of higher and more intellectual end of the financial publishing industry.

The article’s called Should Trend Followers Lower Their Horizons. It’s got some nice equations in it so, you know, spoiler alert there. But it has got a graph showing that in August, when a lot of CTAs didn’t do very well, short-term CTAs did do well. Okay, well, that happens from time to time. You’d expect that, right?

I mean they’re very different styles of trading in many ways. Just because you’re following trends, you would expect the correlation between short-term trend followers and long-term trend followers to be probably not zero or negative, but it’s not going to be especially high because they’re looking for very different dynamics in the market that they’re picking up on.

And there clearly would be times when short-term trend followers will do well, which is when trends are kind of happening, but at short time periods, while at a period of time when long-term trend follows will be whipsawed and will lose money.

And you can imagine a converse situation where a short-term trend follow will be whipsawed because the market’s kind of moving up slowly but in a series of sort of jagged spikes. And each of those spikes is kind of whipsawing the short-term trend followers.

So yeah, it’s fair enough that sometimes short-term will do well, sometimes long-term will do well. But what’s disappointing is the fact that this article kind of seems to imply, to a degree, that this one month of performance, this the one month… It’s not even over the last few years you’ve noticed short-term trend, long-term trend following. It’s one month of performance!

So, another one of my bugbears is people not using enough data to make decisions with.

So, when we’re doing our backtesting of our own strategies, we want to get as much data as possible because we want to get lots of economic cycles, and make sure that we’re not picking up on something that’s just a temporary event. You know, if I had somebody working for me who came up with a backtest that was one month long, they’d be out the door.

You know, the only place that might make sense is in the high frequency space where you can’t get enough data points for the month of data to get a statistically significant result. So, I was quite disappointed by Risk magazine publishing this article because really those guys should know better.

I mean, at the end it does say, well, this is just one month, is it a blip? But that’s after a whole article full of implication that, oh my goodness, August was terrible, therefore we should be trading shorter. So, there we go.

That’s two publications that I respect greatly, The Financial Times and Risk magazine, doing quite a poor job, I’m afraid, with portraying our industry this week.

[00:08:56] Niels

So, clearly, we’re not going to get any love from those magazines anytime soon. But now that you talk about the topic, I did actually notice that as well. I didn’t realize it was from Risk magazine. And I also thought exactly what you explained. I thought that’s a very odd article to come out with, especially because some of the names they mention as having done well for one month, we know that they’ve been struggling for a long time. So, it’s completely nonsense. But that’s what you get sometimes in mainstream media.

[00:09:26] Rob

I’m going to make one more point on that article, if I may. I just noticed that they’ve got a graph which shows (it’s worth looking at the article, it’s not a terrible article), one thing that’s quite interesting in it is that they’ve got a graph of CTA dispersion, by style, showing different categories of CTA and how they did in August. Now, as I said, don’t draw any conclusions from that graph because one month is irrelevant.

What is quite interesting is that if you look at the categories of CTA that they’ve got there, they’ve got trend following (which I guess is sort of is kind of us probably and probably represents the majority of the CTA industry still), they’ve got short-term, then they’ve got option riders, option strategy, multi-strategy, market neutral, and fundamental. A lot of these things don’t sound very CTA to me. But then CTA is one of these nebulous concepts and we can have a long debate. I think I’ve touched on it before, but what does it actually mean? But of course the new category in there, you won’t be surprised to see, is AI.

So, the AI CTAs are now being tracked on this dashboard and no doubt we will see lots of people jumping on that particular bandwagon and that’s another of my bugbears. But anyway, enough, enough.

[00:10:37] Niels

Yeah, we need to save something for next time you’re back. But interesting and I appreciate that.

So, what hit my radar has nothing really to do with CTAs as such (but this is also the segment where we can go a little bit off the beaten path, so to speak). I was thinking, actually, of the selloff in bonds which does relate to CTA performance certainly this month.

It pretty much started when The Fed cut 50 basis points, and which was actually probably more than most people would have expected. But they probably didn’t expect that we were going to get higher rates in the bond markets as a result of that.

Now of course we know that other central banks have continued to cut since. ECB did it earlier this month as well. They’re down to 3.25%, I think, now. And it was the first back-to-back cut in 13 years that they did.

But what’s really interesting, and what sort of reminded me of this was a comment I saw from Paul Tudor Jones, the famous hedge fund manager who spoke on CNBC. And essentially, at least from the clip I saw, he’s saying that whoever wins (and you and I promise not to talk politics, so we’re going to stick to the Paul Tutor Jones part), he said that whoever wins, he fears that there will be kind of a point of recognition where bond vigilantes (and for all I know he could be one of them) may start to vote with their feet and give a vote of no confidence, not so much to the candidate, but actually to the budget plans that either of them really stands for. He shared a chart that shows the US budget deficit could go to as much as 200% of GDP in 30 years if we continue down the path.

Right now, the US debt, according to this interview, is 35 trillion dollars but the income from tax receipts is only 5 trillion per year and the deficit, at the moment, is running at 2 trillion a year. And it continues like that as far as the eye can see, again, according to him.

And, and then he made a reference to something that you may know about. I did not know about this because it’s not a sport that I follow. But he compared it to this wrestling entertainment (it’s not real wrestling of course, it’s very popular, naturally it is). So, apparently, there’s a term called ‘kayfabe’.

I’m not entirely sure what, specifically, it means, but he called it an economic kayfabe where the fact that you’re looking at something, you know it’s an illusion (what you’re watching), but you’re asked to believe that it’s real. And so, he expects kind of a Minsky moment coming at some point. He was asked what about fixed income for him for his own portfolio, and he said, quite emphatically, that he’s not going to own any fixed income and he’ll be shorting the back end. And I think this is super important.

We know that not that many strategies are able to profit from a long-term bear market in bonds, but trend following has actually proven that it can, very recently. So anyways, that caught my attention. Any thoughts?

[00:13:58] Rob

Yeah, lots and lots of thoughts. Yeah. I mean WWE of course is potentially one of the reasons why one of the candidates is sort of in the public eye. I think the first time he was on television was when he appeared at one of those WWE shows. But anyway, we aren’t going to talk about him or his opponent.

I’m reminded of one of these so-called widow maker trades. The widow maker trades, in financial parlance, are where, essentially, it’s a trade that you put it on and the market keeps hitting you with a stick. It looks like a trade that should always be profitable and never is.

One of the great widow maker trades, in the last 30 years, was shorting Japanese government bonds. And the reason it was a widow maker trade was everyone was like, well, this is crazy. The Japanese economy has got very slow growth, it’s got terrible demographics, and as a result there’s a budget deficit, there’s a huge amount of debt. Surely, surely interest rates are going to have to go up, therefore we should be shorting Japanese government bonds.

One of the nice things, of course, about being trend followers is that we could ignore all of that and we could just say, well, these Japanese government bonds are still going up, let’s just buy them. And in fact, I remember very clearly that when I was running the fixed income book at AHL, one of the greatest trends of the time I was running the book was the long in Japanese government bonds.

From memory, it basically went up in pretty much a straight line. And then in February 2013, it sharply gapped down and we closed our position. We were probably long Japanese government bonds for about three years while lots of people were saying, oh, this is insane. I mean, interest rates were basically zero and staying at zero. It was pure carry at that point. But it just kind of kept printing money.

So firstly, I’m glad I’m not in a position where I’m expected to make discretionary calls on the US bond market. Definitely, for sure, I don’t want to do that. That’s a business I think is very difficult.

I think Japan shows us that there are good reasons why a country that, like the US, that can print its own currency and also has a currency that is the currency of the world, and people will be buying US assets and investing in US assets for continuing to the future. I don’t feel that personally… It’s unlikely that we’ll see a situation where US interest rates end up having this Minsky moment. I think it’ll be more like Japan. I think there’s no reason why the US, as a country, can’t support quite a large federal debt, even larger than it’s got now.

The other point I would make is that if you do look at demographics, one of the things that happens as you get an aging society is that people do move more of their investments into fixed income and therefore there is more demand for bonds. And that’ll be another thing that potentially would be keeping interest rates low.

I could be wrong. I could be wrong. But I can see very strong counter arguments against Mr. Tudor Jones, although obviously he’s a man who’s been much better at making discretionary calls in the market than I have.

And I didn’t even know he was still alive, actually.

[00:17:06] Niels

He’s not that old, actually, and he looks pretty well.

[00:17:11] Rob

It just seems like he’s been around a very long time. I think he started quite young.

[00:17:14] Niels

Well, he has, but as you also know then, probably, is that he kind of started his career as a trend follower. So, it’s kind of an interesting in a way. And actually, I think three or four years ago he came out, also on television, and said that if he had to put his money in one strategy (and this is after four or five years where trend following had not done so well), he wanted to put that in trend following. And then, not long after, we had a fantastic run.

My last comment to this, because I don’t disagree with what you said, I don’t disagree with what Paul Tudor Jones said either, but it is interesting. The conversations that I have with investors are very different to the sentiment that he’s expressing, where he’s expressing major concern about fixed income. I see the opposite. People are so excited, and they want to essentially just buy fixed income because now they can get 4% or 5%, they haven’t been able to do that for a while, thinking that’s a great deal. And of course, if Mr. Jones is right, it may not be such a great deal.

We’ll find out. We’ll talk about it again in a few years, I’m sure, Rob.

[00:18:21] Rob

Yeah, and in the interest of transparency, just very quickly, looking at my own book, I am short bunds, but that does seem to be the only sort of fixed income position I have.

[00:18:34] Niels

There we are. You’re like Paul Tudor Jones, you’re shorting the long end.

[00:18:37] Rob

My computer agrees with Paul Trader Jones. At least with German bonds anyway. Not US, though, where I have no position. Anyway.

[00:18:43] Niels

Fair enough. Good stuff. All right, let’s move on to the quick trend following update. But we are going to stick with the fixed income as a theme because that’s probably where trend followers are hurting the most this month despite long exposure becoming smaller in bonds. But they are hurting from the corrections; and equities, of course, as well saw correcting this month, so, all of a sudden. And some currencies are also becoming a little bit more difficult. I guess the only Thing that’s shining this month, really, probably, is gold and silver, at least in terms of the liquid markets. I don’t follow all markets of course. And most other sectors are really kind of mixed to flat.

My own trend barometer finished at 32 last night. I should say we are recording on Thursday this week, so that would be Wednesday evening. That’s on the weak side, suggesting trend followers will be losing money this month. Yesterday, Wednesday, was a down day. So, keep that in mind when I mention the numbers so far because they will be as of Tuesday this week.

BTOP50 down 1.25% but still up 2.88% for the year. Soc Gen CTA Index down 1.83%, up 63 basis points for the year. SocGen Trend down almost 3% this month, now down 71 basis points for the year. The Short Term Traders index is down 62 basis points, flat as a pancake this year.

MSCI World is down 68 basis points but still up 16.67% for the year. The S&P Global Development Sovereign Bond Index, such a long name, down 1.31% this month, but still up 1% for the year. And the S&P 500 up 61 basis points still, as of last night, up 21.5% so far this year.

Any thoughts other than that or should we jump straight in? We have so many questions.

[00:20:37] Rob

Yeah, let’s move on. We’ve got some great questions this week. Thank you to all our listeners for those.

[00:20:41] Niels

Yeah, absolutely, and in no particular order, in fact, I probably should have listed the ones that came in first as the first question. But bear with me, we’ll try and get to all of them.

The first one is from Crypto Captain X3. I think I can safely mention the whole name because it’s not very descriptive. He writes to Rob in this case, “How do you manage intraday adverse price movement risk while running a daily system?

Presumably, if you generate signals based on volatility and return forecast, N times a day, then we have to run dynamic optimization N times a day and then we can use the buffering approach you mentioned.” So, a bit of an insider there. Thoughts?

[00:21:27] Rob

Yeah, well, getting away from the specifics of how my system works, which I might return to in a second.

So, if you’re doing a slow trend following strategy, most people will do their backtesting on daily data. We’re trading sufficiently slowly that the advantages of moving to hourly or even tick data are very marginal. And a really good test to do when you’re doing your backtest on daily data is to see what effect it has. If you delay your fills by one business day or even by two business days and just see what effect that has on your performance.

If the effect on your performance is significant. And by significant, I mean more than say one basis point to Sharpe ratio. So, that’s actually not very much, but it’s basically enough that you’d see a difference, visually on a kind of account curve. Then you should probably think about either getting hourly data (because it does look like your system will be sensitive to delays and you know, potentially delay your execution to the following day or the day after), or it might be that you’re actually trading too quickly. You actually maybe need to sort of slow your system down and look at your trading costs to see if that’s the case.

So, if you do that exercise (and I’ve done it), and you see that there’s no difference at all between executing at today’s close or tomorrow’s close, what that’s telling you is basically that you don’t care about intraday price movements. Essentially you generate a signal.

The way my system works and where a lot of people’s systems work. You generate a signal on a closing price and then you basically, the next day, try and execute the trade that comes out of that. And if the price has moved a lot between where you could have executed the signal and where you actually end up trading the signal. In theory, at least, you should be agnostic towards that.

You know, in a backtest, where potentially prices have moved a lot from one day to the next, you know that, on average, that makes no difference to your performance. Yes, it will introduce a bit of day-to-day noise, clearly if a market has moved a lot. You know, it’ll change the fill price that you expected to get. Were you to take that new price and feed it back into your system, it may give you a different trade than you want to do.

But again, if you’re trading slowly enough, then moving the last price, in a whole series of prices, by even quite a large amount, won’t change the desired position that much, to be honest. So that’s kind of the sort of, you know, the way I would describe it.

Now, my own system is a little bit more sophisticated in that what it actually does is look at the positions it currently has, looks at the positions it wants to have, and then does a dynamic optimization. And what that means is that, in theory, I could actually run that system on an intraday basis and run it every hour or so. And that would be a way of sort of picking up the new prices that are coming into the market, of course, while still potentially having stale prices in markets that are currently closed and just sort of dealing with that in a more sophisticated way. In practice, to be honest, I don’t bother doing that. So, I do, essentially, do this thing and just trading on what yesterday’s close was and then not caring if the price has moved a lot.

[00:24:41] Niels

Do you have any threshold in terms of if you only need to change like one lot or, you know, whatever, you don’t do it until it’s more than one lot?

[00:24:50] Rob

Yeah. So, my system I used to trade had what’s called a buffer in it, where basically, in very rough terms, if the price had changed by 10% of kind of my sort of standard position unit, which is vol adjusted, then if it’s less than 10%, I wouldn’t bother trading. To be honest, because of the size of my account, you know, that doesn’t make that much difference, to be honest. My new system is a bit more sophisticated.

What it does is looks at the overall portfolio and basically says, well, this is the portfolio I’ve got on now. This is the portfolio I’d ideally want to have if I don’t have trading costs. Is it worth moving from here to here, given that trading costs exist?

And basically, you know, that means that it might be that on some days I don’t trade at all because it’s just not worth it. Or if I do trade, it’ll be in markets that are cheaper to trade. So, that’s a sort of more sophisticated way of doing it.

[00:25:41] Niels

Yeah, no, that makes sense. Then Crypto Captain X3 also goes on to say, “Also how to manage overnight gap risk for futures instruments which trade only six hours a day.”

[00:25:54] Rob

Yeah, I kind of answered that. I’m basically saying you don’t care about, you know, in your backtest you essentially said there’s no difference to me executing yesterday’s close and today’s close. So, I also don’t care about, you know, the sort of surprises happening overnight either. I just, just know that, on average, I’m going to be okay.

[00:26:12] Niels

And then final question. “Let’s say I want to trade directional strategies, but instead of using futures, I want to use options: depending on forecast strength, a mixture of short options and long options. For example, if extremely strong cap trend forecast, 50% long options 50% short options. If extremely strong carry forecast, 100% short option. What are the pros and cons of this?”

[00:26:36] Rob

Well, the main con is that options are more complicated than futures. So, from a kind of implementation and infrastructure and data perspective, you know, there’s a lot of work to do.

Generally speaking, if you’re buying options, buying options actually is quite similar to trend following in the sense has a very similar payoff structure, especially for short term trend following. If the market goes in one direction or the other, then you have a sort of nonlinear payoff of whether you’ll make money or not. If the market stays where it is, you’ll get whipsawed and you’ll lose money. So that’s like, kind of like paying an option premium.

So, we can actually treat trend following, short-term trend following, as a bit like buying an option without buying an option. Or we can buy options. And the difference between those two approaches is that, with the explicit buying of an option and buying of a premium, you’re paying what we call the volatility premia. You’re basically paying for the fact that, generally speaking, options are more expensive than they should be given where volatility generally ends up.

So, buying options consistently is a money losing strategy. And you know, the reasons for that are people like buying insurance policy, they like the certainty. If you think about it, if you’re going long in a market, if you do it by buying a call option, then your downside is limited. And that sort of certainty of position is kind of quite a nice thing to have.

And also, if you buy a call option, it sort of does this natural thing of kind of buying more as the price goes up, which is what trend followers tend to do either because of the way the signal’s constructed or because they’re using some kind of form of pyramiding.

The buying of a call option automatically does that for you because as the price goes up, the delta of the option goes up, which means you’ve effectively got more exposure to the market. So, it’s quite nice, in that sense.

So, if I was to replace my futures trading portfolio with a system that just bought options if I thought the price was going to go up, and sold options when the price was going to go down, I would expect my performance to be significantly worse because I’m now, essentially, rather than buying these options implicitly, I’m buying them explicitly. I’m paying the premium, I’m paying the volatility of risk premium to the market.

Now you can be a bit smarter than this. What you can do is say, well, what I’m going to do is I’m going to trade options and futures and I’m going to put on a bet in the most efficient way possible given my forecast for direction and volatility and implied volatility. So that sounds quite complicated, but let’s give a really simple example.

Suppose I think the market’s going to go up, so I know I want to have something that’s going to make me long. Okay, let’s keep it simple. Suppose I have the option of either buying a call option or buying a future. Which of those choices should I select? (Let’s not use the ‘O’ word in a different way.) Which of those choices should I select?

Well, if implied volatility is currently really low, and actually below my forecast for historic volatility, it may make sense to buy the call option because the call option is going to be cheap and that’s going to be an efficient way of getting that risk on. If, conversely, implied volatility is high, and on average this will be the case, then I want to buy that future instead rather than paying this big fat premium for the call option.

Things get a bit harder to get your head around when you also introduce the option to short options. When you do that, what you will find is that you will end up with a system which has an interesting behavior because it’s going to have elements of both trend flowing in it, but also, it’s probably going to be systematically short vol. It’s going to be bias towards selling options because, as I said, generally speaking there’s a volatility risk premium.

So, you’re going to have this interesting strategy where you’ve got the positive skew properties of trend following combined with negative skew properties of options selling. But, in theory, at least, that system will have a higher Sharpe ratio than something that just trend follows using futures because it’s got this additional source of return from option selling and also from being able to tactically buy options when they’re cheap.

[00:30:53] Niels

All right, cool. Moving on to a question from Taylor and actually I think we have answered the question so it’s going to be very quick. But I want to recognize Taylor for writing in. He says, “How do you think about position rebalancing and the tradeoff between maintaining your desired risk allocation versus minimizing transaction costs?”

[00:31:11] Rob

Yeah, so, as I said, very quickly recapping, there are two ways I’ve done this in my career. The way I used to do it was a very simple way where, essentially, if you actually look at the theoretical literature around rebalancing, if you imagine transaction costs were zero, then you’d always rebalance. If your position was out of line with where you wanted to be, you’d always rebalance. Because if transaction cost is zero, there’s no downside to doing that. Introduce transaction costs and things change.

The more transaction costs are higher, the more tolerance you’re going to have from being away from the position that you want. Because you’re basically saying, well, yeah, I want to have this position here, I’m currently in a different place, but actually it’s going to cost me 50 basis points to move over. Well, it’s not really worth doing that. So, I’m going to wait until the position is even more out of line and then at that point I’ll rebalance.

And there’s this idea, depending on the kinds of costs that you have and whether you’re a retailer or an institutional trader, there’s sort of different theoretical ways of doing it (which I won’t go into detail now, but I have talked about in my books and on my blog posts, and people can refer back to that). That’s the way I used to do it.

The way I do it now, as I said, and already alluded to, again, is more sophisticated. And what that does is look at the whole portfolio and it’s basically doing the same thing. Which is, is it worth going from the portfolio I’ve currently got to portfolio I’ve theoretically have in the absence of transaction costs?

And that’s going to result in more nuanced decisions. Like, for example. Yes. Okay, let’s say that I’m currently, topically, currently short bonds and my theoretical portfolio wants to be long bonds. What I’m probably going to do is go long and buy more of bonds that are cheaper to trade and possibly not even touch my… If I’ve got other positions that are really expensive, I may not even touch those.

So, I may end up with a sort of getting my bond exposure through the market that’s currently cheapest for me to trade. So, it can do more sophisticated things like that, which is, you know, less intuitive. But again, theoretically, it is a better way of doing it.

[00:33:23] Niels

A quick question, by the way. Do you have like a ballpark figure of how much total transaction cost commissions you pay every year? Just a ballpark?

[00:33:31] Rob

Yeah, yeah. So, my annual standard deviation target is about 25% and I pay about 1% in transaction costs.

[00:33:40] Niels

Okay.

[00:33:41] Rob

So, it’s sort of 1/25th of my risk target, if you like.

[00:33:46] Niels

Right, yeah. So, I think it might surprise people that it’s not a massive amount – transaction cost. And actually, I think you’re a little bit on the higher end compared to, for example, what we are when we…

[00:33:57] Rob

Well, bear in mind as well, I’m a retail trader and half of those costs are commissions and half spreads. If I was an institutional trader, I’d probably be able to get those commissions down quite dramatically and I’d probably be looking at more like 60 basis points, of which 50 basis points would be spread, 10 basis points would be commissions. So that’s one reason why I’m a bit higher than most institutions.

[00:34:20] Niels

Makes sense, completely, still. Good. All right, a quick question from Richard. There are a few questions, actually, I see.

“Last year I launched a small systematic fund which trades equities and futures using a lot of your ideas and frameworks. Things are going well so far. I’m grateful to have come across your books and blog. If you ever find yourself in Australia, please get in touch.”

I think I should have probably left out that one, but there we are.

“In your book Systematic Trading, you describe position inertia, a method of avoiding frequent small trades by widening the target to 10% above and below the desired position. Have you considered varying the parameter depending on the estimated trading cost of the instrument?”

Now, again, I didn’t read ahead. I should have done that because these are all the same questions.

[00:35:06] Rob

Yeah, well, no, no, no, actually, this is quite interesting because having described the method of position inertia, I can now explain where this 10% figure comes from. And yes, the truth is basically that in theory you should actually, yes, have a wider buffer if an instrument’s more expensive, and a smaller buffer if it’s cheaper. And that just intuitively makes sense.

In practice, however, I’m a simple guy and to me it was just too much like hard work to have a different size buffer. Also, because I’m trading retail size trades and most of my trades are a single lot, you know, the using, say, of a 5% buffer or 20% buffer probably isn’t going to make a lot of difference to me. So, 10% is basically a conservative value that, for the vast majority of instruments that I consider cheap enough to trade, would be the right value. But yes, in theory, and certainly if I was trading institutionally, I would use a buffer that varies, definitely.

[00:36:01] Niels

Here’s a question, actually, from Richard that actually I find interesting to hear the reply to. He’s saying, “I currently use a rolling 10-year window of weekly returns to forecast correlations between futures instruments, as you suggest in your blog post in 2020. Do you think there would be any value in applying an exponential smoothing to the correlation forecast?

[00:36:24] Rob

I really want a link to that blog post.

So, I may have miswritten in the blog post, but let me just quickly explain the two different ways I estimate correlations and why they’re different. It comes down to what the correlations are for.

So, the first thing I might want to estimate correlations for is if I’m looking at the underlying instrument returns themselves. So, you know, what’s the correlation of, say, two things we talked about already with of the bund and with the US 10-year bond? You know, it’s going to be probably like 0.5. What’s the correlation of, say, the bund with say the Schatz (which is a German two-year bond)? It’s much higher. It’s going to be about 0.9 probably.

Now, that I use for things like estimating the risk of my current portfolio, which, as I said, I use using my dynamic position sizing algorithm. But also, it’s the standard way of measuring risk. Because you look at the positions you’ve got on, you look at the correlation of the returns of those positions. That’s sort of standard financial mathematics.

Now for that particular use case, I actually use roughly a six-month window and I do use an exponentially rolling method. So, there are two separate points there. The reason I use six months not 10 years is that yeah, correlations between market instruments do change. Between the Schatz and the Bund? Not so much. But between US 10-year and bund (which is German 10-year), you know, there’s going to be times when the US and German economy are doing quite different things, and their rate cycles are misalign and the correlations will be lower. There’ll be other times when they’re moving in lockstep, probably during crisis periods, because both bunds and US 10-years are seen as safe haven assets. So, you’d expect correlations to go up in crisis periods. That’s kind of a truth about financial markets is that correlations always move to dramatic levels when there’s a crisis, partly because of things like flight to safety and partly because of sort of deleveraging events that are happening across a lot of assets at the same time. Anyway, I’m digressing.

So, I use six months because that is a good balance between being too fast and therefore the correlation is moving around too much, and being too slow and not picking up, essentially, regime shifts. And why should you use exponential weighting rather than non-exponential weighting?

Well, exponential weighting, as sort of this nice property updating with more recent information, than weighting that more heavily the information that’s longer ago, which intuitively makes sense but also is smoother. So, if you use a fixed window. Imagine using a 6 month fixed window, and 5 months and 28 days ago there was a huge return spike, you know, a huge difference in returns between two assets. In two days time, when that kind of rolls out of your six month window, your correlation estimate will sort of change quite sharply, potentially.

Obviously, this is more of a problem for short windows. Six months isn’t too bad, but for short windows is a very serious problem indeed. Using an exponential weighting, that kind of big spike, it will gradually dissipate out of your returns rather than jumping in one go.

So, if I’m measuring the correlation between instrument returns, I definitely don’t use 10 years. I use about six months.

Now let’s suppose that instead of measuring the correlation between the underlying instruments, I want to measure the correlation between the actual trading strategies themselves. So instead of the bund and the US 10-year, I now have a trend following strategy on the bund and a trend following strategy on the US 10-year.

The correlation between those is going to be lower than the correlation between the actual assets themselves for reasons that I’ll actually explain later on in the episode. But more importantly, it’s going to be more stable.

Generally speaking, there won’t be kind of episodes when that correlation is very high or very low. Of course it will vary a lot, but it’s actually more stable over time. And because it’s more stable, I actually want to use as much data as possible. Ten years, you know, whatever, you know, 10 years, 20 years, 30 years, as long as it’s kind of more than one economic cycle, that’s probably long enough assuming you’ve got enough data to do that.

And with a big, long window like that, using exponential weighting, it doesn’t really make a lot of difference. In fact, in my own code I do. But because it’s such a huge, long window, it’s not going to make a lot of difference.

So, I would be curious to know if I’d actually suggested that in a blog post because as I said, the 10-year window I do use but only for correlations between sort of what I call sub strategies for correlation between instrument returns themselves I do use a much shorter window. And yes, exponential smoothing is definitely the way to go.

[00:41:03] Niels

Yeah, I mean, I’m glad we made that clear. And maybe Richard needs to just go and check what he is doing in relationship to that.

Let’s move on to a quick question from Michael. I think this is also a very quick one. He just writes, ”How do you monitor that your systems are running normally intraday without checking the performance?”

[00:41:24] Rob

So, basically, my system publishes a web page every 10 minutes and I can click on it now. Of course, the viewers at home can’t see this and neither can Niel’s but I’m clicking on it now and it’s showing me that all the processes I expect to be running are running. And if any of them crashed, it says ‘crashed’.

And if things have gone really pear shaped, it also shows when it was last updated. If that’s more than 10 minutes old, I know that the whole machine has gone down, or perhaps just this monitoring process has gone down. So, that’s sort of how I keep track of things in J.

And the other thing is that my system sends me emails basically every night saying everything seems to be okay. For example…

[00:42:2] Niels

Sleep well Rob.

[00:42:13] Rob

Sleep well, Rob. Yeah, so I can tell you, for example, that I’ve got what I call a status report that basically checks that all those processes are… So, if I haven’t clicked on the intraday thing for the whole day for some reason, it basically says, oh look, this process is delayed, it should have started, it hasn’t started, or it’s crashed. It also says things like, you know, you haven’t had a price for this for, you know, this Chinese index equity future for two days. That seems unusual.

There may be a good reason. It might be a holiday in China, maybe the exchanges closed. But it may be also that there’s a problem with my data feed. It will say things like, you know, look, normally you generate a position every day for this market, you haven’t generated one. Again, is that an issue with the subsidiary system running?

And it also does things like say, oh look, here are the positions that the broker thinks you have. Here are the positions I think you have. Hang on a second. There’s a difference between them. Most likely what’s happened is that futures expired.

So, I’ve built a whole infrastructure of things around the fact that essentially I’m quite a lazy person. And I’ve got better things to do than sit around watching a computer run. So, I made the computer kind of constantly say to me, yeah, it’s fine, it’s fine, it’s fine, it’s fine. And then suddenly go, oh, it’s not fine, just to make the actual work involved with running the systems be the absolute minimum.

[00:43:38] Niels

Good stuff, good stuff. Now, you may fall into the AI category in the article you talked about earlier.

[00:43:45] Rob

Oh, another one of my bugbears.

[00:43:46] Niels

I’m not going to go down that road now. Now we have two more questions, Rob, but I need very short answers because otherwise we won’t get to the, to the real meat of the conversation today.

Another Richard writes in, “I’ve read Rob’s book on systematic trading and thought it was in the top five books on the subject of all time. He talked a little in there about allocating capital between trading strategies, but the method seemed quite manual. I’m wondering what method he would use to dynamically allocate capital between different strategies in a live system given dozens of strategies, all with changing performance over time.” One minute answer.

[00:44:24] Rob

Well, the short answer is no, I don’t do that. And I don’t do that for two main reasons. One reason is I’ve got, you know, decades and decades of backtests, different strategies. I’m likely to see a statistically significant change in what I think their performance is based on 1, 2, or even I’ve been training for 10 years, 10 years of data. The second reason is that, actually, it’s very hard to see any statistically significant difference in returns.

That means that, generally speaking, when I’m doing any kind of optimization, I usually almost completely ignore the performance of strategies and instruments. And that’s why I can use quite simple techniques. I think when he says manual, that’s code for simplistic. I use quite simplistic techniques that actually work much better than any kind of more sophisticated techniques that are likely to overfit.

I can use those techniques, definitely, but when I use them properly, taking into account sort of statistical significance, I generally find they give me results that are very similar to what doing this very manual, more simplistic way.

[00:45:32] Niels

Sure, sure. Cool. Good stuff. Last question. It’s from, I think his first name is Cloud. Cloud be a ‘her’, I don’t know. “There are some questions I really would like to get answered. One, among the strategies such as momentum, horizon, acceleration, how do you select which strategies? It seems more like strategies are making money on some instrument, but some of them do not. Also, as time goes by, it looks like some of them became nonprofitable. Are there any ways of optimizing to pick the right strategies and instruments in a profitable way?”

[00:46:07] Rob

I just said don’t do that. Yeah, and I mean don’t do it. It’s very, you know, what looks like kind of a clear pattern, when you actually get down and look at the data and do proper statistical tests, you see, actually, the results are purely what you get from random chance. So, I’ve got a very striking graph definitely in my third book. I can’t remember if I repeated in my fourth book.

But what I do is that I get all 100, I get 100 instruments or so, and I rank them according to their backtested performance for a particular set of trend following strategies. You see a very clear picture where there’s an instrument on the left that does really badly, an instrument on the right that does really well.

And just looking at that picture thing, well, goodness me, what I need to do is allocate all my capital to this stuff on the right and ignore this stuff on the left. And then I add error bars. So, I add error bar showing what the amount of statistical noise is and the estimates of those performances. And once you do that, what you basically see is that all of the error bars overlap.

What that’s telling you is that, actually, you don’t know for sure that this thing on the left is really terrible and this thing on the right is really good. And for that reason, you potentially shouldn’t be trying to optimize performance in that way, at least the kind of trading that we’re doing.

So, you know, high frequency traders, lots of data, statistical significance, is an entirely different matter. But for the slow trading systems we run, I just stay away from it.

[00:47:23] Niels

Yeah, absolutely. And again, just think of the Cocoa example where for 15 years it made no money and in the last 18 months it’s made all the money in the world for trend followers. So yeah.

All right, let’s move on to the main event, which are two articles. The first one is from a former colleague, I understand, at AHL, Yoav Git, now at Gresham Research, and it is called Less is More, more or less. Some of you may already have guessed what we’re going to talk about if you’re a longtime listener, but I will let you, Rob, lead the way.

[00:48:03] Rob

Yeah, so this is one of the things, one of the very few things that Niels and I disagree on, I think, which is diversification. So, you know, Niels is currently wearing (listeners can’t see this), he’s currently wearing a t-shirt saying 50 Markets Is Enough for Anybody. Whereas, I have over 300 futures markets in my own portfolio. So, this is quite an interesting take on it. So Yoav, he’s a very interesting guy, very original thinker.

He’s also much better at maths than I am, and he comes up with these little things that I would never think of just mainly because his math is better. And he’s an original thinker. But what he basically does in this article is talk about the idea that returns in portfolios are driven by factors.

Think about stocks, the stock risk factor, which we often proxy with the S&P 500. And there’s this idea of beta where you say, what’s the exposure of a stock to this factor? And the original capital asset pricing theory says, well, basically all stock returns should be explained by beta. There shouldn’t be anything else on top of that. And the idea is that managers try and outperform by getting alpha on top of the beta. That’s where the term alpha comes from. It comes from this capital asset pricing model equation.

But he does something quite interesting which is to say, well, as you add, make your portfolio bigger, mathematically you actually get less diversification because you end up with less idiosyncratic to stock risk and you end up with more, effectively, exposure to the underlying stock factor, the S&P 500. And I found this very interesting because I’m a big proponent of using equal rather than market cap weights in long only portfolios.

Because basically, by using a market cap weight, you’re saying that you think that say Apple or Nvidia are worth having sort of 50 times more in your portfolio than whatever the smallest S&P 500 stock is. When I last checked, I think it was Gap, the clothing company, but it’s probably changed now. Whereas with equal weight it’s just saying, well actually I think that Nvidia and Gap are about to do as well as each other.

Momentum changes that slightly because obviously once a stock starts doing well, its weight will increase in a market cap weighted portfolio and you’ll end up with more and more of it. If the stock does badly, its weight will reduce or maybe even get kicked out. So, momentum actually makes market cap weighting look better. But in the absence of any information about stock performance, you should have equal weights.

But what Yoav shows is that even if you’ve got equal weights, as you add more and more stocks to your portfolio, basically it becomes more and more essentially like the S&P 500, even if you’ve used equal weights rather than market cap weight. So that was kind of interesting to me.

Now the other thing he then does in the article is just take this back to CTAs. Now, again, we can think of CTA performance as being explained by some factors. Now it’s less obvious what those factors are, and they may change over time, but I like to think that there are probably four, maybe five factors out there that potentially explain CTA performance. And, generally speaking, from time to time you can clearly see what’s going on. Like a couple of years ago we had the big movement in commodities and there are times, go back to say 2008, where it was a very clear risk off bet with stocks, selling off bonds going up, and that explained basically CTA performance very clearly.

So, the factors may be different over time, and change, but generally speaking there aren’t that many big things going on. And as a result, if you do the same argument essentially of sort of saying (like I say), you know, we should have 300 futures markets, what he said is, well, actually you end up with this strange situation where you end up with the portfolio becoming more and more loaded towards a series of risk factors.

So, you may think you’ve got 300 bets, well, actually you’ve got five bets. Because your portfolio is so big, the contribution of each of these individual futures and the contribution of those idiosyncratic risks (so, something like cocoa, that’s completely off the wall), clearly, if you’ve got 300 markets and cocoa’s only got one of them, well that’s not going to make much difference to what’s driving your performance. What’s going to be driving your performance is these big kind of macro trends that are moving lots of markets together.

So, I found it thought provoking, definitely. I know this, for example, if I look at the increase in performance in a backtest of a trend following portfolio as I add markets to it, it does increase by a factor of about four to five, which is brilliant. I mean there aren’t many things you can do that will bring you a fivefold improvement in performance.

But what that’s also telling you is that, actually, there are probably only about (for technical reasons) it’s the square, four or five squared factors, individual kind of independent bets, that are actually driving your performance. And of those, between 16 and 25, probably only half a dozen are actually important at any given time.

Now I have to say that, you know, of course Yoav is a smart guy, but like a lot of us, he’s working in the industry. So, he’s publishing this because he has a CTA that’s different from what other people do. And I won’t go into details about what exactly it’s doing, but he’s saying basically, well, you should be looking for things that have got lower correlation.

So, if you’re just a plain vanilla trend filling CTA, there’s probably no point in having 300 markets, you know, unless you can do something very different with those or go down the route of even like the alt market push. So. the movement by people like Florin Court, into alternative assets, even that kind of only buys you a little bit more diversification. You might be introducing that one new factor into your model, but really it’s, it’s quite limited, yeah.

[00:54:20] Niels

The way I think about it, in a very simple way, and I don’t have the math skills of any of you two, is that people often talk about, yeah, we can introduce more… And by the way, it’s not that I don’t think more markets will improve, it’s just that I’m not in a 300 market camp saying that it’s better. It’s different, I just don’t think it’s better.

Anyways, the way I think about it is that often when you hear this narrative, you hear it as, well, if I trade 300 markets, there are clearly more chances for me to find that one or two or three markets that really have a big trend, as you rightly pointed out. As you also rightly pointed out, well, the position size is most likely going to be quite small. So, it doesn’t make a huge difference really. And it may even be worse, actually, because the position size is so small.

But what is often not talked about is the fact, yeah, if you add another 200 markets to your portfolio, you’re also adding the chance of 200 markets being stuck in a range and you losing money on those. So, this is kind of just my simple reasoning for why it’s hard to argue that it’s better.

And then as I often say, and people will probably have heard this many times, when I look at all the performance databases I can find, and have a really fairly good guess in terms of who’s trading 300, 400 markets and who are trading, say 100 or less. There is no evidence, in my opinion, that one is better than the other. And that’s all I would say about it. But it’s different.

And different can be sometimes, you know, useful, but I don’t think it’s better.

[00:56:03] Rob

Yeah, it’s worth saying as well that you have links to an episode that Katie did earlier this year, in March, as well, where she talks about how this relates to skew. Because if you think that there’s sort of these latent macro factors which should drive, you know, trend following factors that are out there, then actually for the best skew, you want as pure and exposure to those factors as possible.

So, it’s one of these classic things in finance. There’s no perfect answer, you know, between what you’re getting, adding more markets may well improve your skew, which actually is counterintuitive in a way because you would assume that your skew will come from having these little independent bets that have going to go massively to the moon, as they say in the crypto land, and provide that big right tail.

But actually, as the portfolio becomes bigger, it’s the exposure to the underlying macro factor with its positive skew that’s doing that; the little independent idiosyncratic bets that, if anything, actually because the tails, could be left and right and sort of wash out in the end.

So yeah, it’s interesting and I think what’s potentially interesting, for someone who’s sort of wearing an allocator’s hat, is that he’s sort of arguing, well, maybe people allocating to CTAs should be thinking a little bit more like people in stocks do. Which is you have your kind of core beta allocation which would probably be a fairly vanilla, low cost, you know, CTA. Then you’d have something a bit more interesting to where you think there’s potentially alpha and lower correlation. Now at the moment, I guess… It’s an interesting argument.

I guess most, most people and most funds are in the business of doing both, right? They’re kind of selling you both at the same time in the same way that, if you go back along a long time, if you bought into an equity hedge fund, a lot of their returns were coming from beta and they wouldn’t sell you the alpha separately, even though that’s what you really wanted and they would charge you fees on the whole thing.

I do remember a fashion, about 15 years ago, and AHL participated in it as well, to launch a kind of simple low-cost fund with lower fees as a beta product. And I remember saying at the time, maybe this is good from an investor perspective but from a business perspective it’s kind of crazy. Because there’s a big risk this could outperform the flagship fund and people would be going, oh, are we paying you guys, you know, 2 and 20 when there’s this thing here for 50 basis points flat, you know. And so yeah, it’s an interesting argument.

The only caveat I would say, and yeah, I would agree with this, I think, is that, you know, this is all linear, right? Correlation is a linear measure of returns and if we know anything it’s that financial returns are not linear and the correlations potentially probably understate the amount of diversification that’s available as you add more assets, especially in trend following. And definitely that’s what my own research shows me.

If I look purely at correlation, the sort of diversification benefits should level off much earlier than they actually seem to do because there does seem to be these kind of nonlinear extra returns coming in from as you add more of these assets.

[00:59:25] Niels

Maybe we saved the best for last, I don’t know. Maybe in terms of your pet peeve, we certainly have saved the best for last, I think.

Now, we are going to talk about replication because a new paper came out from the guys over at Newfound Research, Corey Hoffstein’s team, and they all always are worth listening to and reading what they write. The paper is called In Pursuit of Trend Following Beta: the Promise and Pitfalls of Replication.

I think people know some of my views, maybe not all of them. And, of course, I speak about this with Andrew quite often. But I don’t know if we’ve talked about it before, but I’d love to hear your thoughts and guide us through the key takeaway points. We probably have about 10 minutes to talk about it.

[01:00:19] Rob

So, the reason I wanted to talk about this paper second is actually one reason why replication should work is that if there really are just a small number of macro trend factors out there, and all the markets were just kind of, you know, effectively just living off those if you like. Their replication should make a lot of sense because what you basically want to do is use the data to find those four or five factors. And then, in theory, you could build a portfolio that kind of replicates them and, obviously, without paying CTA fees.

[01:00:56] Niels

Which are not 2 and 20 by the way.

[01:00:58] Rob

Which are not 2 and 20. Now this was, like I said, this is 15 years ago, unfortunately. In the good old days. Actually, I used to charge even more than that for some customers, but we won’t go into that now. So that’s a reason why replication should work.

And actually, the way I trade myself (I’ve described it briefly), is effectively a form of replication in that there is a portfolio I cannot trade because I have a small capital account, relatively small account, that consists of positions in 300 instruments. And then actually on a day-to-day basis I’m probably trading about 20. Those 20 are effectively a replicating portfolio of the big portfolio.

But that’s replication based on a position level replication. So, there are, kind of very quickly, three ways of doing replication. Okay, position level replication, which is where you know what the positions need to be and then you form a series of positions that get as close to that portfolio as you possibly can.

The second way of doing it is through what I’d call a sort of bottom-up approach where essentially you say, well, I know what trend followers do. They do things like moving averages. They do things like position sizing. Why don’t I just build a really simple model that does that in, say, 30 liquid futures markets and that’s going to be much cheaper and more straightforward than another whole kind of business of trading hundreds of markets and doing much more complicated things?

And the third way of doing it is basically through what I could call a return replication. And what you do here is you start off with a series of returns that you want to replicate, which is normally but not exclusively an index. So, like the BTOP50 or the SG index. Basically, what you do is you say, well, I’ve got these futures and what I want to do is essentially to produce a series of long and short positions that, in the recent past, would have produced the performance that I see in the BTOP50 or the CTA index. So, these are the three main approaches.

My biggest bugbear is with the third kind of approach. And I’ll probably touch on the reasons why. The first kind of approach, I use it myself. Clearly, I have no problems with it. The second kind of approach I also think is very valid and a very good way of getting a kind of low-cost CTA into the hands of people who particularly… I think it’s really good for something like an ETF structure where you’re trying to reach people who haven’t got the capital to put into a managed account, for example, or to reach a hedge fund minimum. So, it’s really good for that. The third one, which is where a lot of the marketing and the noise has been around and where this article is around, the return-based replication is the one I have a problem with.

But anyway, let’s talk about the paper. The paper’s really interesting because it does something that I think is really good, which is rather than using real data, it uses random data, randomly generated data. And this is a brilliant approach that I think quants should use more because it reduces the chances of overfitting.

What it’s essentially doing is saying, well look, if we look at replication based on real data, it’s worked really well, but that might just be a fluke, okay? We don’t know whether that just so happens because of the way trends have worked out in the last 30 years, that you could do replication very easily.

So, instead, we’re going to use random data. We’re going to basically randomly generate a bunch of pseudo CTAs and a pseudo CTA index and we can repeat that experiment over and over again and we can see how well replication does. So that approach itself is really good and it’s an approach that’s really useful.

And actually, there’s a lot of kind of useful thought and color that comes out of that. So, for example, the dispersion we see between different CTAs that have quite high correlations, if you look at the returns over long periods on a year-to-year basis, it reproduces that effect. So, it reproduces the fact that in some years CTAs have got a big dispersion in performance. In other years they have a small dispersion in performance.

And if we go back to this idea of latent risk factors driving performance, then what that’s telling you is that in years with low dispersion, probably there were one or two big factors that, you know, everyone made their money out of. In years with high dispersion, there were a lot of weird things going on in different markets and depending on what you know, exactly how people trade those markets, what the market’s weights were, they’d get quite different performance. So that’s really interesting.

The article then goes on to talk about the sort of replication process. What they basically do is say, well, given this artificial universe, we’ve got these artificial CTAs, how well does replication do? And, you know, it does very well. And they kind of find that sort of the first two factors drive a lot of performance in this sort of pseudo universe.

And the first say 30 factors basically explain everything. And therefore, you only need about 30 futures as long as you choose your universe to have a reasonable amount of independence. So, if your 30 futures are all US bond futures, then I’m not sure if there are 30 US bond futures. But, you know, if they were all the sort of the same thing pretty much, then of course replication would do very badly. You need to have a good spread across asset classes, and for various reasons you’d want to potentially choose more liquid instruments as well.

And they have this interesting calculation, that I’ve never seen before, that’s really interesting, which is basically to say, well, what’s the amount of variability caused by replication versus the index? And what return can we expect purely through cost saving?

And they said, well, the return is like a return and the variability is like a standard deviation. Hey boys, what we got here is a Sharpe ratio.

So, if you divide the potential saving through reducing costs by the kind of tracking error of the index, they say, well, we get a Sharpe ratio of close to one.

[01:06:58] Niels

Can I just comment on that particular part because it was something I noticed as well. And what goes into that cost is 200 basis points of management fees (as I said earlier, I don’t believe that’s a good number anymore), 110 basis points for performance fees (that could be true, depending on what the long-term performance is), and then 275 basis points of unnetted transaction costs. And I’m kind of thinking that compared to the cost we pay, compared to the cost you just mentioned, this seems high.

[01:07:33] Rob

It does seem high. And just to make it clear, the reason for this phrase ‘unnetted’ is essentially, if you imagine a situation where you’ve got lots and lots of managers independently doing slightly different trades, but they’re all going to be paying transaction costs. If you had them under a single roof, you could actually get them to net-off a lot of those trades, and as a result reduce the transaction.

[01:07:51] Niels

How would you be able to calculate that?

Anyways, I’m just saying I did notice that number and I just want to say that to me it’s a little bit of a stretch.

[01:07:58] Rob

Yeah, yeah, it is a bit high. And to be fair, I think what they’re trying to do there is say, well look, that’s basically a ceiling. You’re not going to get any better than this with replication because, A, actually, the variability comes from an in-sample fit. Okay, so clearly there’s problems with that. Which we’ll look at in a second. Secondly, you know, that’s like at the top end of the savings.

We can, we can dig down on numbers and so on and so forth, but it’s probably not unreasonable to be able to make an extra 100, maybe 200 basis points through a combination of a little bit of transaction cost saving and a bit of fee saving. That’s assuming that whoever’s doing this isn’t charging any fees themselves, of course. So, you’re going to have to take that off. But, but you know, it’s fair enough.

So, the other thing they do is then they actually do a… That’s sort of, as I said, that’s a ceiling. When they actually do this exercise on the BTOP50, they find that they can’t get to that near-one sort of Sharpe ratio, if you like. They get down to 0.23. So, that’s you know, 0.23. So, we’ve gone from 600 basis points down to about a quarter of that, so about 150.

So, you know, essentially that proves your point, Niels. The savings aren’t really that large but you know, it’s still 150, maybe 200 basis points. It’s still better than a kick in the teeth.

The other thing they do is they use… So, the problem with return base replication is that you’ve got a series of returns, they’re probably monthly, and you’re doing a regression. And the regression window is… We talked earlier about correlations, about how long your data period should be to estimate a correlation. What you’re trying to capture here is the effect of what happens when people’s position changes sign. So, you can’t…

Yeah, go on Niels.

[01:09:57] Niels

I just want to say that you said the returns are probably monthly. I’m pretty sure they’re daily in order to make the regression.

[01:10:02] Rob

Yeah, yeah, yeah. So, they’re potentially daily. But what you’re trying to do… So, you couldn’t use even a six month regression because the problem is that during that six months the position… Definitely the size of the positions have changed but the sign of the positions has probably changed as well.

So, you may do this regression, say over six months or a year or whatever and say, well, it looks like we’ve got a positive low to US 10-year bonds, which is like saying that CTA managers are long 10 year-bonds. Well, yeah, maybe there were over the last six months of the last year, but not now, probably. So, you’ve got to reduce the size of your window to capture that. Which means the number of data points you use has to be lower.

But the things you’re regressing against, on the right-hand side of your regression, you need to have quite a few of them to get some… Let’s say they’ve got 30 instruments, 30 futures (liquid futures), you’re trading because you need to have a reasonable spread, like I said. So, for technical reasons, if you’ve got a lot of instruments on your right-hand side and not many data points, regression becomes very difficult. And in fact, you get to the point where you theoretically cannot do it.

So, to deal with that they use something called Lasso, which is essentially a regularization technique for people who are into that kind of thing. And what that essentially does is say, well, you’ve got 30 futures, but actually, to explain the returns over the last three months or however long you’re regressing for, you really only need five of them to achieve your regression.

So, it’s a nice paper. It has some interesting insights and uses a couple of novel techniques that I think should be used more. The main one being this idea of random data. Lasso is used a fair amount. That’s not something so special. But to me at least, and I’ve read this quite reasonably carefully, so I could be wrong, and if I am, I apologize, but it still looks to me like they’re doing an in-sample test.

So, I think a proper test of this methodology is basically something where, you know, you do say your regression, and then you form your portfolio of positions that you have, and then you say, well, how well does that do? Let’s say you’re doing this regression every week. So, for that following week, you know, you hold the positions as they are. How well does that do? And then tracking the portfolio. And we know that out-of-sample never does as well as in-sample.

That’s why backtesting is such a fraught business. So, to me, they’re doing a lot of work to make something that, fundamentally, I think is a bad idea a little bit less bad. Okay. And we have an expression for this, in English, which is called ‘polishing a turd’.

I feel like there’s a lot of turd polishing going on here because to me, at least, there are much simpler ways. If you want to get a simple, cheap way of replicating CTAs, you shouldn’t be obsessed about trying to get a close match to, say, the BTOP50 index.

I think that people who are obsessed with benchmarking might think that’s a good idea. But if you really just want to get some simple, cheap trend following, well, you know, just do 30 futures in a very simple portfolio and just bung that into an ETF. To me, that’s a much more sensible thing. And what that will do is automatically capture the change of sign as we go, say, from being short bonds to long bonds or vice versa. It automatically captures that because you’re going to have a little momentum model inside there that’s doing that for you rather than trying to kind of back it out from the returns of the BTOP50.

[01:13:30] Niels

Now, you may have read this more carefully than I have. I take some of your points. Of course, we’re going to hear more about replication. As I mentioned earlier, these are smart guys, so we need to pay attention.

There’s one thing, and this is from memory, where I think it focuses a little bit on this tracking error. And I think they come up with a number of 6% or 7% or 8% (I can’t remember what it is) as a tracking error, and then say, well, it’s not much worse than the tracking error, or it’s more or less the same as a tracking error of, of managers, as far as I remember. But of course, managers are not trying to meet a benchmark, so I think the concept of tracking error, as such, is not so relevant.

And one thing I have noticed with the replication products that are live, that we know of today, that Andrew represents for example, as far as I remember, there is quite a lot of tracking error. Frankly, the correlation is not super high to the actual benchmark, at least not on daily data, as far as I remember. So, I’m probably more in your camp. Are you trying to replicate it with high correlation or you’re just trying to beat the benchmark with, you know, a couple of percent because you think it’s cheaper to run it the way you do?

I think the narrative is a little bit mixed up, but maybe from a sales point of view, from a marketing point of view, it makes sense to say, yeah, I can save you a couple of percent by doing replication. But I’m not entirely sure that they truly replicate, and I guess that’s my question, do they truly replicate the performance?

I think Harold, from Trans Trend, wrote kind of a response to maybe some of the narrative where he said, no, they’re definitely not because they’re not giving you all the diversification you get in the individual managers and so on and so forth. But I’m still questioning this in my own mind. What exactly is it they are delivering? Because if it’s a true replicator, I would have expected the correlation to be higher, to be able to claim that you’re replicating something.

[01:15:44] Rob

Well, as I said, I think for me, fundamentally it’s not possible to do a good replication job with a purely backward looking measure based only on returns. I think that’s a fundamentally flawed way of doing it. I can understand why, from a marketing perspective, you say, well, we’re going to give you the benchmark performance, less the fees paid by the underlying managers. Isn’t that a wonderful thing? You know, at the end of the day, as an individual person, what are you going to do?

You’ve got a choice in lots of options, right? So, you could, for example, if you’ve got enough money, pick say 10, 15 CTAs and dump, you know, a chunk of money into them. You’d need to be quite wealthy to do that because, you know, assuming you’ve got say an industry standard allocation of 5% or maybe 10% CTAs and you then want to allocate, you know, to 10 of them, well, you’re looking at 1% of your assets being a big enough ticket for them to accept. So that’s an option available. Not to me certainly, and to very few people even listen to this podcast, I suspect.

Or you could say, well, I think I can pick a good manager, and you invest in a single manager. Or you might say, well, I’m going to pick this manager. They’re probably okay, they’re doing something like trend following. The fees aren’t too bad. But I know I’m probably going to have dispersion to an index, right? Because they’re in a single manager and by construction, even if they’re included in that index, they’re going to have dispersion to it. You know, something just like the replication people can.

Or you might say, well, what I really want to do is have a… Of course, we could argue that it’s misguided. So, Yoav’s argument is that it’s misguided really to do any of this. Right? And so, you might as well just pick a manager and try to pick one that’s doing something different. Because actually, if you invest in lots of managers, or even in one manager with a diversified portfolio, you’re just getting more and more exposure to a small number of risk factors. And if you genuinely believe that’s the case, well, actually something like a bottom-up replication, done cheaply, may make sense to you.

It may make sense to buy something like an ETF that has, say, 30 futures markets in it, and runs a very simple momentum model on them, and only charges you, I don’t know, 50 basis points for the privilege. If you believe that there are just a small number of risk factors driving performance and that the benefits diversification over lots of instruments, lots of managers are limited, then do that.

But trying to get the benefits of diversifying across lots of managers via the BTOP50 index, via a return-based replication process, I just think it’s a bad move, I really do. So, anyway.

[01:18:28] Niels

My final thoughts are, I think what some of this has created is, of course, especially in the ETF world, I think it’s created an avenue for people who would not otherwise be able to get maybe an allocation. But this was before.

Some of the managers now are offering some version of their strategies in ETF format, which obviously is another way to go. But anyway, it certainly attracted some assets from maybe smaller investors. I think that’s a great thing. So, I’m all for that.

I think for institutional investors to say, let’s just go the replication route, first of all, I question why would they even have that job if they’re just selecting a replicator? Because they must believe that, within managers, there is something called skill, and they must believe that they are themselves, skilled enough to find managers that can, over time, persistently, you know, create skill. I’m not talking about, you know, every single year that they outperform, but over time there are managers that we know that outperform the benchmark, so to speak.

So yeah, Rob, I appreciate all the preparation you did to dig into all this information. So, I’m sure a lot of people would have enjoyed that. And if you did, please show your appreciation, go to your podcast platform of choice, iTunes, Spotify, Amazon, leave a rating and review for the podcast and thank Rob for another great conversation.

Next week it’ll be Alan sitting in for me and he’ll be speaking actually to Andrew. I have a feeling they’re going to be a little bit of a rebuttal to what we talked about, Rob, but we have no control over that.

[01:20:13] Rob

I look forward to hearing it.

[01:20:14] Niels

If you have a question for Andrew or for Alan for that matter, you can always send them to info@toptradersunplugged.com and I will pass them on to Alan.

Anyways, that’s it for now. From Rob and me, thanks ever so much for listening. We look forward to being back with you next week. And in the meantime, take care of yourself and take care of each other.

[01:20:38] Ending

Thanks for listening to the Systematic Investor Podcast series. If you enjoy 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.

And remember, all the discussion that we have about investment performance is about the past, and past performance does not guarantee or even inform 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.