249 Systematic Investor Series ft. Mark Rzepczynski – June 24th, 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] Niels

Welcome and welcome back to this week’s edition of the Systematic Investor series with Mark Rzepczynski and I, Niels Kaastrup-Larsen, where each week, we take the pulse of the global market through the lens of a rules-based investor. Mark, wonderful to be back with you this week. How are things? How are you doing?

[00:00:38] Mark

Very good. It’s the beginning of summer. This is officially the first weekend of summer, albeit in the US, we usually start on Memorial Day and go to Labor Day. But from an astrological basis, this is the first weekend of summer.

[00:00:53] Niels

I have to say, here in Switzerland, it’s been feeling like summer for a while. It’s been very warm and humid, but I’m sure that it’s been the same on your side.

Now, Mark, we have great number of topics that we’re going to talk about as usual, a long list that we need to tackle. But before we do that, I’m always curious what’s been on your radar since we last spoke a few weeks ago.

[00:01:20] Mark

Well, the main thing we’re going to talk about is in the main topic because I could go on and on. It is just what has been happening in the first half of the year or what hasn’t happened or what we expected to happen that didn’t happen. That’s probably what’s on my mind right now that all of the crisis issues that I thought were going to be important at the beginning of the year haven’t turned out that way.

[00:01:44] Niels

Fair enough. That’s a good little teaser to what we’re going to talk about. I’m not sure whether this is something we expected to happen or not, but I will just say that as I was preparing our topics today, this morning, on Saturday morning, there are of course many, many breaking news everywhere about a possible coup d’etat in Russia. Who knows what the real situation is? But of course, if that is the case, then there will be some interesting market action no doubt Monday morning.

Now, despite it being a relatively quiet week in terms of news, that is until this morning of course, it was a week of corrections in many of the trends that we’ve been enjoying in the last couple of months. In particular, we saw this week equity markets taking a breather, and that led to some giveback in some of the open profits for trend following strategies. Another market that kind of saw the same pattern this week was sugar. That also saw a little bit of a price cut, so to speak, against it this long-term uptrend.

Elsewhere, I think it was pretty mixed the way I see it at least, even within the same sector, in some cases, for example, in metals. I think you had markets like aluminum, nickel, zinc doing probably relatively well and then all the other metals probably not so well this week. The same could be said about fixed income where I think some markets like the short SONIA, what used to be the short Sterling in the UK, had a pretty good week for those who still have short exposure, but then other markets saw small losses as they were moving higher this week.

So perhaps the only sector overall that was pretty congruent and still doing what it’s supposed to do from a trend following point of view, and that’s energy, where we did see some lower prices which I think benefited trend followers even though overall, it was a down week, I’m pretty sure of that when everything is counted as a Friday night. In terms of my own trend barometer, still stuck in neutral. In fact, right now, it’s stuck completely neutral at the level of 45.

So if people go to the website and they see the chart over the last 12 months, the trend barometer has really not shown lots of willingness either up or down, which completely matches what’s happened performance-wise so far this year. Because when I look at that right now, I see that the BTOP 50 Index, yeah, it’s making 65 basis points as of Thursday. Friday was probably a down day and it’s down about 1% for the year, so really going nowhere. Same with the CTA Index, up 60 basis points so far in June, down 70 basis points for the year.

The Trend Index also up about 65, 64 basis points, down 1.25% for the year. Then maybe, again, a little bit to my surprise perhaps, the Short-Term Traders Index not really making much of this more, how should I say it, range trading environment. It’s flat for the month of June and it’s still down more than 3% so far in 2023. Equities on the other hand still doing well in June. Up 3.6% for the MSCI World, up 11.5% for the year. The World Government Bond is down about 10 basis points so far in June, while the S&P 500 is up about 4% in June, up 13.25% for the year, of course, as we all know, driven by only a few big tech stocks.

Okay. Now, before we dive into the topics that you brought along, you and I exchanged a little article that came up during this week, mainly, I guess I called it by looking at that title. It was on LinkedIn in that I saw it. It was written by a guy called Sam van de Schootbrugge, as far as I can tell, who’s linked to the Macro Hive group over at Bilal’s team, and he called it Are Momentum Models Profitable?

Now, since we’re in the trend following, space that captures our attention a little bit. So I’m always curious when these people write about something that’s at least related to what we do. However, you are much better than I am to dissect it a little bit for a few minutes to see what they found out and what some of the arguments were. So perhaps I could ask you, Mark, just to tell us a little bit about this article.

[00:06:23] Mark

Sure. The article was associated with something called momentum gap. They said that if you look at the momentum gap, which we’ll define in just a second, that that’s a good indicator to tell you whether momentum strategies are going to be successful or not. Let’s go back to give a little bit of context. A trend follower will always look at markets as a sense of a time series. So you look at each individual market separately, you look at the past time series, and then you determine whether you want to be a buyer or seller.

Momentum, it takes a very different view. What they’ll do is to look at past returns, and usually it may be nine months in the past, and then what they’ll do is they’ll look at all of the returns in a set of stocks cross-sectionally. So instead of a time series, we’re going to look at any given time, we’re going to rank order all of the stocks from the highest momentum to the lowest momentum. We don’t really care whether it’s positive or negative, we’re just going to rank order them. Then what you do is, say for good momentum strategy, and this is no different than the way you look at other factors in the marketplace, you will buy the top of that ranking and then you’ll sell the bottom of the ranking.

If you’re buying the top of the momentum ranking, then you’re saying that whatever past returns will then continue into the future, which is traditional trend following. Those at the bottom, if they were negative, those past negative returns will continue into the future because you’re buying the top and selling the bottom and you’re setting up a market-neutral portfolio that ballances off the two and you’re going to make money based on the top stocks doing better than the bottom stocks.

Now, what we find out is that momentum is a core component of all markets. If you look at the very, very long run, the momentum risk premium or strategy is very successful over the long run, but it has some very high variation. There’ll be periods where it doesn’t make money and then there’s other periods where it makes a lot of money. So what this researcher did when he was looking at momentum gap, he said, well, let’s look at the difference between the highest performing stocks when we sort them versus the lowest performing. That difference, we’ll call as the gap or the range or the dispersion in returns.

So when you think about it, your first view might be if the dispersion is really great, that’s good. In some sense, I want to buy large dispersion or a big momentum gap. I think that that’s intuitively what people would think. What he finds is just the opposite. When that momentum gap gets very large, that actually, future performance will decline. Now, it’s really important how you set up a momentum test. So usually when you do momentum trading, you’ll start at the beginning of the month, you’ll do the sort and then you’ll hold the stocks for the month, and then next month, you’ll resort and do it all over again.

So he finds that this momentum gap or the negative impact of momentum gap, if let’s say you hold the same portfolio for a longer period of time, the negative performance is very strong. If you rebalance every month, it’s still there. It may not be as strong. So what he’s saying is that this is an anomaly because he said it isn’t associated with arbitrage. It doesn’t seem to be a behavioral story. It doesn’t fit within, let’s say, a marginal utility story.

He’s scratching his head and said, I found this anomaly, the momentum gap, and it’s very strong. It not only exists in the United States, but if you looked at 21 countries, I think the number that he found was 15 is that there’s a negative relationship. The other is that it’s certainly not positive in the other direction. So he’s saying we look at in-sample, out of sample, we look at different periods of time, this momentum gap exists and it’s strong. So he said this is this is an important result.

Then I started looking at this and said the one thing that’s missing when I look at this is, how do trends behave over time in general and how would momentum? So if we have a large momentum gap, that means that we’ve already had big performance spread between the best-performing and the bottom-performing. If let’s say on a go-forward basis, you make less money or your returns are lower, then it suggests that if there’s a big dispersion between high-performers and low-performers, that won’t continue.

The reality is that markets are mean reverting at some point. So in some sense, this gets out to the number one issue that we see with all trend following, that at some point, it’s one issue to try to get into a trade, but the real important issue is determining when to get out of a trade. What this researcher has found with the momentum gap is here’s a good reason to get out of trades when the dispersion gets very large. So it’s interesting. This actually addresses the longer-term issue of trend following. I probably say there’s always two types of people. So you’ve always got to set up straw men of one extreme versus another.

There are some trend followers that would say trends lasts longer than expected so you should always stay in a trend because you never know how long they last. So that would be the one school of thought. Since you don’t know when they’re going to end, you just got to write them out because you’ll always be surprised. The other school of thought would probably say that I’m a trend pessimist. A trend pessimist would say, I’m going to be a trend but I know that at some point, it’s going to reverse because everything in the world mean reverts.

Herb Stein said that something that cannot go on forever won’t. So you’re always paranoid about being your trend follower because you say that there’s always going to be a time when that market is going to revert, because if it didn’t, then it would take up all of the assets of all the money all over the world. So this really talks to the issue of markets eventually will mean revert. So what do we do? What indicators can we have to determine whether markets are stretched?

Some people use the word dispersion, I like the idea of stretch. That’s consistent with this whole idea of thinking about change points. If markets stretch and they’re like rubber bands, eventually, they’re going to have to snap back at some point. Now, if you try to get out of trends early, you will be disappointed by some trades that go on for much longer than expected. So in some sense, there is this behavioral aspect of are you comfortable with the fact that you may be leaving money on the table because there may be some trends that will last longer than expected or do you want to take some of your profits early?

This is consistent with all of the prospect theory that we’ve talked about in the past and which I’ve written about that generally, people like to sell their winners and hang on to their losers. Now, in some sense, when we talk about the utility or the behavior of people who sell their winners and hang on to their losers, we always talk about well, of course, that’s irrational. Yet, on some sense, if you believe that markets revert, that at some point, you want to say you’ve got to take some profits or you want to think about taking profits.

But this gets at the heart of the issue of prospect theory, behavioral finance, stretch, and the issue of do trends lasts longer than expected. So I think this was a very interesting paper. It’s causing me to think, is there another way I could look at or create an indicator within my market set to think about this as an indicator of stretch? Now, the interesting part is that I, as someone who looks at markets and follows trends, said, oh, this seems obvious.

The researcher said, and I may be exaggerating, he’s saying I’m perplexed by this answer. I’m not sure why this would occur when in reality, it seems a little bit obvious that when markets are in extremes, then they’re probably going to snap back.

[00:16:48] Niels

I agree. By the way, thanks for that explanation. I agree with you. it’s always been my own gut feel that (and some people will disagree with the words I choose) where I’ve said in the past that I thought that the exits were more important in defining how successful you’re going to be. What I meant by that is not to say that the entry isn’t important. Of course, it is because if you don’t get into the trade, you’re not going to make any money. Of course. So we know that.

But my gut feel is, and I haven’t done the research in detail, but my gut feeling is that regardless of what kind of trend following system you choose to use, if it’s more or less set at the same speed, it’s going to get you in more or less at the same time, whether it’s time series, momentum breakout, moving average crossover, whatever it might be. So I’ve always felt that entries were somehow “easier” to come up with. But as you say, what really defines how much of the trend you can capture is your exit. At least that’s how I see it.

Now, I have come across, even in our CTA series, there is a manager or two that uses that kind of stretch concept, I believe, which I think most trend followers wouldn’t really go along with because, to some extent, that’s what I would consider a profit target because it takes profit even as the market continues to move in your favor. When I think of exits, I think they are generally some market trades that are triggered by some kind of move against the direction of your position.

So you’re not selling into a rally. You are selling because it’s peaked and now it’s going down. Then it’s just a question of how sensitive your exit strategy is, etc. But it is interesting. Of course, there are many ways of doing these things. There isn’t one way.

[00:19:01] Mark

The exit is so critical on all these trades. Let’s give some simple examples. If I put a trade on, fairly quickly, you’re going to know whether it’s going to work or not. I’m not saying that you always want to use stops, but let’s say if you had stops, if it’s not working right away, you’re going to get stopped out and you’re going to get nicked. You’re going to give up some profits, but that’s it, and then you go on to the next thing. Or if let’s say that prices start to move above a moving average and then it reverses back, boom, you reverse the trade. You get nicked, you’re out. I don’t want to say that’s easy, but your risk that you take is contained.

The problem that you often find and I’ve had internal arguments with people on this issue is let’s say I have a trade and I put on a trade of corn. Let’s say I put it at $6 a bushel, then it goes to $6.30. Then all of a sudden, it starts to reverse and then you finally get out at $6.05. Some people would thump their chest and say, well, I started out at $6, I made $6.05, and therefore, it was a very profitable trade and so look, I’m a really good trader. I, being the pessimist (pessimists are not always gloomy), but I would say, wait a minute, if the price went up to $6.30 and I got out at $6.05 and I started out at $6, basically, I gave back more profits than I actually generated at the end of the trade. That’s a lousy trade. I can’t believe how stupid I was that I didn’t get out at around $6.30.

I’d be kicking myself and saying there’s something wrong with my model because I wrote down losses that were much greater than what I actually took as profits. Now, some would say, Mark, that’s just the cost of doing business, just get over it. I would probably say that that’s why the exit is so important because if somehow I could figure out how to get out, assuming I can’t get out at the peak, but if I could get out at something better than $6.05, I would have had a much better trade.

Always ask yourself that, well, let’s look at the hump. The hump was at $6.30. If let’s say I got out at $6.25, then the next person did the same trade but he got out at $6.25 but on the other side of the hump. Are those the same trades? Were we both successful? Was the first trade, the one that got out before we reached the peak and then redeployed the capital somewhere else, was that a better trade? Or was it the same to say, I went from $6.25, I went to $6.30 and back to $6.25? Okay, well, that’s the same trade. It took up more time, it used more capital, I gave back profits. So those are two different trades.

When you think about exit, you’d say, in a perfect world, I’d like to get out before that peak, I’d like to be able to sort of say then I redeploy that capital in a trade that might have higher, better returns someplace else, as opposed to riding it past the peak and losing money from the maximum level.

[00:22:59] Niels

Yes. Now, I don’t want to get in too much of discussion, because actually, this is not even one of our topics. I’m not so sure, I would say that I agree when you say I would rather get out before the peak because I still think that the strength of what we do is the fact that we don’t know where the peak is going to be. So let’s just run with it. I’m not quite in that camp, I have to say.

Let’s move on to the topic that you mentioned at the very beginning. I think this is a topic that we could have said exactly the same thing last year when the war in Ukraine broke out. I think there are a lot of people that expected certain things to pan out. They didn’t. So when you wrote to me and said, why don’t we talk about all the things that did not happen this year, I thought, yeah, why not, let’s dive in. So tell us more. Mark?

[00:23:57] Mark

Well, we will hit the end of the month before your next call, so this is a good time to start to do a recap. But eventually, what we’re going to do is we’re going to end up with something about how this applies to trading. At the end of the day, yes, this is a recurring theme of forecasters who say, I’ve made a forecast, here’s what I thought, and you go back over the forecast and you find out it’s not close to what was reality.

So let’s look about what didn’t happen, how that impinged trend following and CTAs in general and trading, and then why the forecaster is different than a trend follower. So let’s do the big setup. Recession: weren’t we expecting a recession? Now they’re saying it might be 2024 or not ever going to happen. We had so far this half year, a banking crisis. We thought that the banking system was going to come to an end and this is going to be the big opportunity for a big dislocation. That didn’t pan out.

Inflation: we’re stuck at around the 4% core. Now, it’s looking as though the things that- central banks have said if we ever get above our target, we could easily just change a knob and get it back down to target. Now we’re looking at targets that it might take us two years to get there. So what happened there? China recovery: we thought that once the COVID restrictions were lifted in China, that this would be the engine of global growth. That’s not happening. We now have a Fed pause but we’re still baked in about two 25 basis point changes.

The world was coming to an end with a debt ceiling crisis, and then lo and behold, it was almost as though it became, as some people would say, a nothing burger. In a sense, nothing happened. So you look at, what are the trend followers looking for? What would divergent traders look for? They’re looking for dislocations. Yet, if you look back through 2023, we haven’t had the dislocations. So now we look at what the market is. October of last year, we had a volatility of around 30 plus for the VIX. Now we’re at about 13.

When you think about it, a volatility of 13, I’m really surprised at that because everyone would have thought that even at higher inflation, higher inflation would just cause more uncertainty and that would just keep volatility for stocks higher. Nope, that’s not the case. We’re probably at the lower end of the VIX range even though it is highly skewed. We’ve got double-digit returns for the stock market, albeit it’s highly concentrated. So that is still a surprise.

The dollar is probably down 10% from the peak. But given that we’ve paused and given other central banks are also raising rates, that’s not that surprising. We’ll say 2022 was an aberration. You look at commodity markets, oil prices have been highly volatile, but they’re somewhat range bound. If you look overall, a lot of commodity markets are somewhat range bound. So given all of the potential crises we have, we’ve had less trends, less returns, less dispersion, there have been less divergences.

Convergent, I can’t really say we’re in a convergent environment. But we’ll say that for trading perspective, if you say the traditional CTAs are more beta timing managers; they’re looking at overall markets and trying to time the beta. We haven’t had a big beta timing environment. We look at also idiosyncratic risk for stocks and looking at a very short-term horizon, it’s probably been more of an idiosyncratic market.

You look at the combination of stocks within the S&P 500, in the short run, there’s probably better opportunities. It’s not so much that they’re better but they at least exist where the beta timing or large macro CTA hasn’t been finding opportunities. So where are we going? What you find out is that there’s two types of people again. We’ve got the forecast guys.

The forecast guy would say, I have an idea and oftentimes, my risk is getting in early. Their problem is always being early. So I’ve got to forecast the phenomena. Their risk is always on the timing of getting in early. Whereas Jim Grant said, the inevitable may be certain, but is not punctual. So that’s the problem. We may still have a recession. It’s just that the timing of whether it’s going to be punctual or not is the problem. We may have another banking crisis. It’s maybe inevitable given all this. We may have a real estate crisis. It may be inevitable, but the timing is the problem.

Now, someone who’s a trend follower is just the opposite. He doesn’t ask what will happen. He doesn’t try to think about what are the potential events that might occur that might lead to changes in the markets. He just says, it’s just a given. I don’t know or I can’t forecast those events. I’m just going to assume that it’s going to happen. If anything, their biggest problem is that they might be late as opposed to early. So the forecaster is always saying, I’ve got to get in early, and then they usually lose money because they’re ahead of the crowd. They’re smarter than everybody else. I don’t want to say the market is dumb, but the market is usually slower to react to some events.

The trend follower on the other hand says, I can’t tell you what will happen. I’m not going to tell you when it’s going to happen and I’m probably going to be late to the party. So my biggest problem is the event, has it already taken place? Am I too late? So when you think about it, they’re very different approaches on how to think about it. Now, while we say that they are very different, they’re not mirror images of each other. So the forecaster could be wrong. But if the forecaster is wrong, that doesn’t mean that the trend follower will make money. But it’s a different way in which you look at the environment.

[00:31:27] Niels

Yeah, absolutely. There’s a lot of things that tie together with that. So we talked a little bit about, well, you also have something called mind the gap, momentum gap.

[00:31:40] Mark

Going back to momentum gap, I always think about when I’m in London, when you’re in the tube, and they say, “Mind the gap.” So you always have to say let’s make sure that we mind the gaps, because that’ll make a big difference on what your performance might be.

[00:32:00] Niels

Yeah, absolutely. All right, good. Then we kind of jump a little bit into ML, AI, maybe, in terms of topic, something that you talked about, and in a sense, also related to the fact that we live in a time of uncertainty, which, to some extent, I guess we always do. So how do you see this challenge?

[00:32:29] Mark

Well, this has been a very interesting challenge because it used to be that the challenge was, okay, what’s the data I use, and then I try to apply my models to a given set of data and you try to train it or you look at different periods, and then you make some adjustments and try to look at different features to what you’re trying to do. But the ML revolution has really increased the choices that a modeler has. So now it isn’t a matter of say, what rules do I use or what dataset do I apply to or what’s my portfolio construction, but it’s also the choice of what are the tools I’m going to use to try to extract signals from.

In a very simple world, in the early days (and this applies to any systematic model) it was very digital. So, trading or signals were an on-off switch. Let’s take a simple example. If price is above a moving average, I’m going to be long. If price is below the moving average, I’ll be short. There was very little, I don’t want to say flexibility, but…

[00:33:51] Niels

It’s a bit binary, right?

[00:33:52] Mark

Yeah, it’s a bit binary. That’s why we say it’s digital. It’s a switch. We turn the switch on, we turn the switch off. You might have used linear regression as a tool, and that might help you on to have these binary switches. You switch on and switch off. But now, what happens is that we’re well beyond regression. So we have all of these different techniques that apply. All of them are trying to do the same thing, extract signals. But now, the value added for any systematic manager is the type of technique or what technique you have, and now your choices of techniques are much larger.

So it’s not a matter of how you can… well, it’s always a matter of how you construct a portfolio and data use and markets you trade. But now one of the key determinations of saying what’s the difference between one systematic manager and another is the techniques that they use. Those could be neural nets, random forest, gradient boosted. So the list is endless on how you look at it. Now, I think this is all an improvement. But what it does is it puts a whole different burden on researchers and different burden on a quantitative analyst.

It used to be you had one tool or your toolbox was very limited. Now we have a much bigger toolbox, more tools. We have to decide what is the right tool for the right problem and how do we choose that tool to answer the specific problem? That’s not always easy.

[00:35:49] Niels

In some ways, you could argue having more tools is better. It gives me more choice to solve the problem. But you could also ask, well, sometimes having more tools makes it harder to choose which tool to use. So all these possibilities, and Alan and I touched on this last week a little bit about AI and implications for our industry. I’m not sure how I feel about it myself. I don’t know if it’s going to make a huge difference to long-term trend following, maybe short-term trading, perhaps.

I have not personally heard of anyone in particular, except for maybe one of the managers we talked about or talked to in our CTA series that said they use machine learning, but they do it under very strict supervision. So I don’t know. I was just curious, Mark. You’ve also seen many different iterations of systems building. How do you feel about it? Are you excited?

[00:36:50] Mark

I’m excited about it but then it also puts a lot of work and burden on me because now I’ve got to figure out what all these tools are. Every time a new tool comes out, or a new process, someone says, this new process is great. Usually, what you’ll do is you’ll read a paper or there’ll be someone who writes about it, and they’ll say, I took this dataset, I ran with this new tool, and look, I’ve got these good results. Right away, your answer is okay, I better figure out what this tool is.

So now you’ve got to spend the energy to say, what is the tool? Then you have to unpack and say, if I use this tool, am I going to get a better answer than if I use a simpler tool? Is this marginally improvement and what is the marginal improvement that I’m actually getting? Does it lead to more trading or not, which means it could cost more? My general view has always been that we use the tools we want like we use a car. It’s a mode of transportation to get from point A to point B.

If I have a more sophisticated tool or car, well, there’s more likelihood that something might break. If something might break, there’s a potential point of failure. So the example is remember the old Volkswagen Bug, where you had the engine in the back. If something went wrong, you could get a little stool, you open up the back of the car. You probably could figure it out yourself if you’re mechanically inclined to a degree.

So now I could have a Ferrari, but there’s no way that if let’s say something breaks and I’m on the side of the road, I’m going to have to call for a tow truck and probably can have a different response. We’ll say we’ve had this problem in our firm, and it’s not a problem. We use machine learning. We think that it has value, especially if we’re looking at a lot of features and a very broad stock index. So if I’m trying to look at trying to tease out fundamental data with macro data and technical data, and I’m looking at, let’s say, the Russell 1000 or the S&P 500 and I want to look at 500 stocks and I want to look over a long period of time, using machine learning can be very, very helpful.

On the macro side where I’m looking at intermediate to longer-term type of trades, then there might be only marginal improvement or there may not be as much improvement. So it’s also a matter of using the right technique at the right time. We did have an interesting issue that we used some of the more complex features. So we were using some analysis with it and we were using some random forest. At the end of the day, we came back and said, well, let’s look at actually more simpler techniques and those would actually be better.

So I think that all these new techniques, what are they trying to do? They’re trying to tease out nonlinearities inside markets because markets are nonlinear. If we wanted to be more simplistic, there’s always going to be breakouts in markets. They’re going to be in a range, and all of a sudden, they break out. That’s a nonlinear system. So if you try to use linear regression in a nonlinear system, you’re going to have a problem. So using these new techniques to try to get at the nonlinearity.

The other thing it tries to do is to try to make some of the decision-making you do from trading to move it from a digital signal to an analog decision. What we talked about initially is that the digital decision is binary. Yes or no? So an analog would say, well, it’s more of a continual process. So that’s what we want to think of in terms of probabilities. If you think about this also within change points, the probabilities make a lot of sense. So when I look at price above a moving average, we want to say, what’s the probability that that will continue?

It may be that as that price above moving average gets to a certain extreme, then the probability that the next step will be higher may actually be lower. So from an analog perspective, I’d say the probabilities that this trade will continue to be positive is lower so therefore I might take it off at this time. In a digital world, you say, well, as long as price is above the moving average, I just have to hold. In an analog world, you could think in terms more of probabilities and you could then be able to get more nuanced in how you trade. That might be helpful from a machine learning perspective.

[00:42:15] Niels

It certainly will be very interesting to follow this evolution that we’re going through right now and certainly is happening quite fast. I take a lot of comfort in the fact that trend following, as a strategy, has been around so long that we’ve seen many of these, quote-unquote “revolutions”, the introduction of the internet and the whole tech revolution, and so on, so forth. I think it’s the underlying methodology that I firmly believe in and the robustness of the approach and the fact that it’s based on human behavior. Whether or not we can improve using AI, I don’t know. Time will tell. So it’ll be exciting, without a doubt.

Now, Mark, just in the interest of time, we have another 15 minutes or so. So I’ll kind of let you choose of the remaining topics that you mentioned or brought along that you wanted to talk about, which ones you feel might fit our conversation today.

[00:43:18] Mark

The issue I think that’s probably most relevant for a lot of listeners is this issue of failure of models. The reason why I bring this up is, obviously, we’ve been building some new models and we always want to see what’s the performance in sample and out of sample. There were some interesting research papers that talked about the failure of finance in this effect that models fail. What you find is that you test a model in sample, and then all of a sudden, out of sample, it doesn’t work as well. So that’s one problem.

Another problem is that someone writes a research paper and it shows that they found some risk premium anomaly, then as soon as it gets published, the anomaly goes away which might be arbitraged away. But what you find out is that what you find in the lab doesn’t match what happens in reality. So it’s an interesting study where a group of researchers, they looked at a lot of these risk premium models, and then they said, let’s all train the models and then let’s look at what happens out of sample and find out how much decline in Sharpe ratio occurs. The rule of thumb is probably about 30% to 40% degradation out of sample.

I probably would say when I was on the investment side and someone showed me a new model, the rule of thumb would make me say like, well, let’s do this count some backtesting by 50%. That may not be a right number but it’s sort of like a rule of thumb that a lot of people have used. So I think that overall, the number one challenge for researchers is trying to figure out how to make sure that we minimize model failure so that we can say that the difference between in-sample versus out-of-sample performance actually matches up as closely as possible. It’s especially the case if you’re a new manager.

So I’ve probably been spending a lot of time to think about two areas. One is market frictions, which we’ll define in just a second. One of my favorite terms that you know that I’ve used before is VUCA.

[00:45:53] Niels

It’s back.

[00:45:55] Mark

VUCA is back. I love it because it’s used as a framework. The first is market frictions. When you build a model, did you account properly for transaction costs? So models performed really good in sample, but did you also include all the transaction costs or liquidity costs? So if I want to know whether a model is going to fail or how much of a discount I should give, I go right to the transaction costs or liquidity. So can I be able to trade these things? Can I be able to have good transaction costs assumptions?

That’s why if you’re a modeler, if you start trading, you want to look at your model and run it parallel against actual trading, and then say, how much slippage between theory and actuality occurs. If that slippage is too great, then what am I missing from my model so I could close that gap? It’s well known. Everybody knows this is the case. But I think for anybody who’s probably listening, this transaction liquidity issue or the frictions issue is something that you’ve got to constantly look at because if you don’t, then you’re going to be surprised.

The other when I look at the VUCA issue, let’s look at this in terms of model building. One is that when you think of volatility, did you train it in one period and has there been a regime change? So you have to ask yourself that if you’re training low volatility and we have high volatility, you’re going to have a problem. So you want to make sure your training set includes enough phenomena so that then you’re going to be able to handle different regimes.

Then you want to say, in terms of complexity, let’s look at overfitting. So you want to try to keep models as simple as possible. So you’re constantly looking at complexity may be good, but then as you add more and more complexity, you might be maximizing your returns for a given trading period. Then you want to look at the training in different periods to just say, do I have a long enough period or sample size for anything I do?

I know we’re discussing a little bit and this is probably a longer topic, but if you have a long term trend following model, you’re going to need a longer trend period because you’re just going to have less trades. If I have a very short-term model, well, then I’m probably going to have a lot more trades for any sample period. Of course, then you have the other problem that I have more trades for the sample period, but then is it regime specific. So you have a trade-off. I’ll get more regimes for my longer-term horizon, and that’s what you want to have. But at the same time, if you have a longer-term system, you’re going to have less trades so your sample has to be all the longer.

Finally, parameter uncertainty is that how do you deal with the fact that you chose a set of parameters for your model, and then as you move forward, you don’t know whether that’s the actual best set of parameters. So there’s ambiguity about whether what you picked is the right number. So I probably would say that the failure of models problem is the biggest challenge for any quantitative analyst more so than when you’re saying, okay, did I pick the right entry or exit?

Let’s say you have two models, and one has a Sharpe ratio in sample of 2 during the training period. I look outside sample and then it has a performance of a Sharpe of 1. I have a second model that might have a Sharpe ratio around 1.25, 1.10. Then out of sample, it’s around 1, just slightly under. Which one would you prefer? One person might say, well, I always want to choose the high Sharpe ratio and will just say that the out-of-sample return is just bad luck. But it’s not immediately obvious.

If I have a lower Sharpe ratio model but the performance is very stable out of sample, you might say that’s a better model than if it was a high Sharpe ratio but then out of sample, it’s performance has been cut in half. Now, it always reminds me of a story of walking in Manhattan. I was going from one allocator to another. So the first allocator I go to, he says, “If you don’t have a Sharpe ratio of 2 or better, I’m not even going to talk to you, just get out of my office. I’m only looking at high Sharpe ratio strategies.” So I said, okay.

I walked down to the next allocator. He said, “If you’re showing me a Sharpe ratio above 2, I’m going to kick you out of my office because I don’t believe it can do this. It’s going to mean revert and so as soon as I invest, it’s going to be much lower than what you say it is. So I would never invest in a Sharpe ratio above 2.” It doesn’t deal with the failure of models, but it does suggest that there are many ways to interpret what is a good or bad model. But at the very least, you want to make sure that your training performance is similar to your actual performance.

[00:52:21] Niels

Yeah, this is so critical and it’s probably one of the hardest things we do, and also identifying when perhaps a model is not doing what it’s supposed to do. So I have a lot of respect for those people working in research and dealing with these types of challenges. I actually also have a lot of sympathy for allocators having to make an outside call, essentially, because there are certain things we as managers can’t share on these things, and then going through drawdowns and having to decide whether this is just part of the path or whether it’s something more, a sign of something not working, especially when you’re outside must be incredibly difficult.

I think that’s where communication from managers is super, super important. Frankly, it really comes down to something quite basic to some extent, and that is, we as managers sometimes really need to hold our investor’s hand through periods where we are pretty convinced that this is part and parcel of how the strategy operates. Of course, sometimes these things happen, I wouldn’t say at times where other managers are not necessarily going through a difficult time, because there is obviously a certain level of correlation. But sometimes we will look different.

One of the questions we always get from potential investors is, how are you different? Of course, we don’t know exactly what other people are doing so we have to explain how we think we’re different. Then people might make a choice based on, oh, yeah, I like something like this because it’s different. But then when you are different during a time and maybe you’re underperforming and other managers are performing really well, then people suddenly don’t really like that difference, even though it’s the difference that originally attracted them to your strategy. We’re dealing with human beings where things are not static in terms of how we view things. Yeah. So this is fascinating stuff.

[00:54:39] Mark

It also applies to existing managers. So I’m talking about the problem for you have a model, you’re showing the model and it’s in-sample versus out-of-sample training versus actual performance. I’m sure you’ve seen this. You could be a long-term manager and yet you’ll have periods where you’re going to be doing well and then there’s going to be periods that you’re not going to do as well. If you’re an investor, you constantly have to ask, is the future performance going to be, if maybe they are going through a drawdown or have poor performance for the last couple of years, do I expect that the last couple of years is going to be representative of the next few years?

If I look at the longer-term track record, they might have a much higher Sharpe ratio and say, well, yes, this is a little bit of an anomaly and I’m going to go on the long-term record. So this is not just between training and out-of-sample. This is for anything we do. I think there’s also you can’t discount the fact that there’s such a thing as good luck and bad luck. So if I run a simulation or I bootstrap, I’m taking one path. So I could have the same model but I could have alternative realities of different paths of return.

I assure you that when I look at those different paths, I always pick the good path. But we don’t know that that is actually going to occur. There’s almost a cloud of where your performance might end up and there is good and bad luck with models.

[00:56:27] Niels

Yeah, absolutely. Now, we’re coming up to the hour level. Anything you wanted to round off with any particular of the last few points that you think we definitely need to just touch on before we wrap up?

[00:56:43] Mark

Well, I think that as we look at this half-year mark and we go into the new year, and this is consistent with my idea of macro awareness. It doesn’t mean that I’m a forecaster, but I say you want to be environmentally aware. We’ve had surprisingly low macro uncertainty that has led to low market volatility. But if you look at the second half of the year, if you bet on the distribution of the VIX, you bet on the distribution of what is the macro uncertainty, the expectation is that the second half of the year is not going to be like the first half.

But my bet would be that we’re probably going to have more volatility. We might even find out on Monday after the things in Russia, but there’s no there’s no question that we had a surprise to the downside and lack of macro uncertainty, a lack of market volatility. So a surprise on the downside, not that we’re going to mean reverting guys, sort of, but it’s more likely that we’re going to be on the upside and we’re going to have more dispersion in volatility in the second half of the year.

[00:58:09] Niels

Yeah, I think this is actually one of the things that I find fascinating at the moment, and that is, when you look at the geopolitical perspective of what’s going on in the world. If you also look at the macroeconomics, what’s going on, and the massive changes we’ve had in interest rates, the massive changes we’ve had in inflation, it really is kind of surprising.

Then you look at a VIX that during the week, I don’t know if it closed below 13 or it was certainly during the week below 13, and you think, wow, that is a low level of uncertainty priced into those markets. It is surprising, and I think this is, again, why I think what we do in our world is pretty robust, because often when people make these calls about, oh, yeah, we’re going to have this and this environment over the next six months, and often, it just doesn’t pan out. I think even I’ve been surprised how complacent some of these markets are behaving given all the uncertainty that I think we have somewhere beneath the surface.

[00:59:26] Mark

It’s interesting that everyone will agree that the stock market is a predictive market, so it’s forward-looking. But because you’re forward-looking doesn’t mean that you get there your forward-looking forecast correct. The thing that always sticks in my mind is a very interesting study by Niall Ferguson, the historian. So he looked at the stock market prior to the beginning of World War One, so after the shooting, after armies started to mobilize but there really weren’t hostilities. So he said, let’s look at the volatility of stocks.

Let’s look at what’s happening in the stock market. Did the stock market predict these geopolitical unrests? The answer was no. It was really surprised by World War One. That doesn’t mean it’ll apply in everything. But there can be large events that occur and markets could still be complacent even though they’re predictive and forward-looking. So that’s the thing that always should scare you.

[01:00:35] Niels

Yeah. All right. Well, we don’t want to get anyone really today, But I appreciate all of these comments, thoughts, insights. As people will know who’ve been listening for a while, we try to touch on different angles of these things. We’re not really necessarily always saying what people should do, we just want people to think about some of these topics and make up their own minds. Of course, we have our own biases, we have our own beliefs, but for me, it’s very important to have this as an ongoing conversation about topics that I think are important.

At the end of the day, it’s about finding something that you as an investor can really believe in. Because if you don’t believe in whatever strategy you end up with, it’s going to be very hard to stick with it for the long term. I think most investors will say, or at least most successful investors will say that it truly is something that you need to apply for the long term in order to achieve your goals, whatever they may be. So we’ll keep doing this, Mark, bring up topics for people to think about. So I really do appreciate all of this today.

Of course, as always, in order to broaden our listenership, we would love if you would give us a little bit of help by going to your favorite podcast platform, leave a rating and review. It really does help our podcast reach a bigger audience. Of course, always happy if you would share the episodes with your friends and colleagues. Next week, speaking of complacency, we’re going back to the world of volatility, because Cem will be back on the Systematic Investor series. So we’ll see what his thoughts are.

He had some strong opinions last time he was on about what the markets were going to do and actually was pretty right that they may surprise to the upside before and it was before the markets were going to have some weakness in them. Let’s see where he is on his thoughts on that. If you have a question for Cem, you can email it to info@toptraderunplugged.com. I’ll do my best to bring it in front of him next weekend. Actually, you should send it before because I think we’re recording on a Friday.

Anyways, you can of course also go and check out all the blog posts were posting on a weekly basis, bi-weekly basis actually. There’s a new monthly trend following update that Rich and I put out this week. So hopefully, lots of things for you to do this weekend. From Mark and me, thanks ever so much for listening. We look forward to being back with you next week. Until next time, as usual, I hope you will take care of yourself and take care of each other.

[01:03:45] 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.