273 Systematic Investor Series ft. Richard Brennan – December 9th, 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 Richard Brennan and I, Niels Kaastrup-Larsen, where each week, we take the pulse of the global markets through the lens of a rules-based investor.

Rich, it is wonderful to be back with you this week. How are you doing down under?

[00:00:38] Richard

Very good, Niels, but it’s very hot. It’s decided to turn on the sunshine in December down here in Australia, so it’s boiling here at the moment. But I’ve got all of the fans going and I hope it cools down a bit.

[00:00:52] Niels

Yeah. Well, actually, I’m also in a place right now where it usually is pretty hot, even in December. But there’s a nice cool breeze coming through these days, so I’m not struggling as much as you are.

Rich, as usual, before we dive into a question or two that we have, and also a very interesting topic that you brought along, I just wanted to find out a little bit of what’s been on your radar recently. What you’re finding interesting when you look at the world at large.

[00:01:23] Richard

Well, what I found was that I had a pretty good result for November, and that’s because my portfolio doesn’t have much bond exposure. So, I came out of November a bit cocky thinking that I’d done quite well in comparison to a lot of the other CTAs that might have had a bit of difficulty because the bond was over the month of November.

So, here I was, a bit cocky at the end of November. Then the first week of December erupted on my screens, and yes, my beautiful commodities, which had offered these great trends and my yen pairs, were absolutely hammered.

Fortunately, Friday gave me a bit of a respite. But yes, this month might not be the stellar month that I’m hoping for.

[00:02:10] Niels

I think we all felt, looking at the screens yesterday, “thank God it’s Friday”, and the unemployment numbers came out a little bit different than people expected. Speaking of that, I did notice yesterday, in the unemployment report, that the unemployment rate of 3.7% marked the 22nd straight month of US employment or unemployment, I should say, below 4%. That’s the longest streak since the 1960s.

By comparison, the unemployment rate never once fell below 4% in the glory days of the 1980s and 1990s when everything was supposedly much easier. So, it’s interesting to see what the numbers are telling us and how people are feeling about the current economic environment, which I’m not sure people think is such a great one at this point.

[00:03:02] Richard

Agreed. It’s a tough environment at the moment. It’s very uncertain as far as I’m concerned.

[00:03:08] Niels

All right. Well, let’s leave the trials and tribulations for December until the end of the month so we know what the final score is. But speaking of trend following, of course, as we normally do, you already touched upon it a little bit, it has been an interesting start to December.

Certainly, fixed income markets have been challenging for the longer-term trend followers. Maybe some of the shorter-term models have already turned and have done well. But of course, then again, the number yesterday may suggest that perhaps it’s a little bit too early to completely conclude that the Fed is done.

Also, I was looking at a chart for just one-week price movements, and you also mentioned that there were some very large percentage changes. Not in isolation, but actually, if you look across the board, markets generally seem to be releasing a bit of volatility at the moment. Of course, not least spurred by the biggest one-day rally in almost a year on Thursday after the Japanese monetary authority offered a surprisingly clear hint at a shift in policy inflicting some pain, (to say the least), on short positions in the yen and also, maybe, long positions in Japanese equities.

But there were also some other commodities that didn’t do so well last week from a trend following perspective. Sugar, I think, was one of them. You may know a few others. So, we have started December off on a soft note, but we’ll see how it all ends up. Clearly, it’s not going to be the same as last year where CTAs, in general, were having a great year. This has been much more challenging.

I guess this is perhaps our last conversation in 2023, except for the group conversation, Rich. So, if you have any thoughts on the year as a whole, I would love to hear your thoughts.

[00:05:11] Richard

Well, after the prior year, this year has certainly been a year of stagnation for me. Early in the year, I had some positive results. But over the course of the year, it’s been quite a challenging year for me. I might come out of this 12-month period with a slight increase in portfolio value. But apart from that, it’s just been one of stagnation for me. But with this change of trend that’s occurred now, you never know where that might lead. This early December change might be signaling something coming.

That’s the thing about this game we play, Niels. We just never know. I was hoping that December might give me a bit of a more solid result for the end of the year, but I don’t necessarily think it might translate that way. So probably, at best, I might come out of here at breakeven.

[00:06:10] Niels

Yeah. Now, I think you’re absolutely right. What I’ve noticed this year is that my trend barometer has really been stuck in a very neutral to negative range for almost all year. Very few observations were above 50 in contrast to last year where we had quite a few of them.

So, I think all in all, it’s confirmed the markets haven’t been as trending, but I imagine that you’ve enjoyed the recent Bitcoin rally. I think you do trade Bitcoin, don’t you?

[00:06:39] Richard

Yes, I have, Niels. That’s probably one of the only bright sparks on my portfolio, at the moment, the Bitcoin. I was hoping that there’d be a bit of a good solid gold move, but I had a nasty little spike this week and it doesn’t seem to have translated into much.

[00:06:53] Niels

Quite a huge move in gold, actually. I think it was like down $200 in one day.

From a performance point of view, this is as of Thursday evening. Actually, when you look at the indices, I will say it really doesn’t look very bad so far this year. The BTOP 50 is down 83 basis points in December. It’s only down 1.81% for the year, and it was in comparison up almost 15% last year. SocGen CTA down 84 basis points, down 3.5% for the year, but it was up 20% last year.

The Trend Index down about 1% in December, down 4.7% year to date. It was up more than 27% last year. The Short-Term traders Index had a little bit of a comeback, up 44 basis points this month and down 1.79% for the year, and that was up 11.27% last year. MSCI World, of course, is doing well, up 78 basis points, up 17% for the year. World government bonds are also still doing well, up 89 basis points. Finally, the S&P 500 Total Return Index of 83 basis points and up 21.8% this year.

So, with that in mind, we’re going to take a couple of questions from Adam for Rich, before we jump into your topics. Let me see what Adam wrote here. He wrote, “A couple of questions for Rich next time he’s on the show. I hear the phrase ‘loose pants’ quite a lot, which makes sense. My question is how loose is too loose, with regards to the distance between the entry price and the initial stop? I see an obvious trade-off in that the tighter initial stops will result in a lower win rate, however, larger profit multiples on the winning trades. Whereas wider initial stops may result in improved win rate, i.e. not getting stopped out unnecessarily, however, lower profit multiples on winning trades. As such, how do you strike a balance without overfitting the data to ensure the winners produce enough profits to pay for the many losers? I feel backtesting this is the simple solution, however, I worry that this may result in curve fitting it to the past. From experience, is there a typical range of ATR multiples that appear more reasonable over the long term, say i.e. 2 to 3 ATRs or say 10 plus ATRs?”

What are your thoughts on this, Rich?

[00:09:38] Richard

It was a great question by Adam. Just to put this into context, Adam is right. The way I deal with this is in the backtest environment. The way I do it is that I’m determining my stop distance by applying my models over very large intensive datasets. I’m not just applying it over one market, because my models are referred to as being universal models because it’s targeting this feature called outliers, which are universally present within any liquid market.

It’s not looking for specific properties of the markets, it’s looking for these tail events that occur in these markets. So, the way I do it, therefore is I might have a 30-year history on 100 markets, and by undertaking my process of testing my models, I’m identifying the optimal stop loss across that entire selection of markets. So, if you could imagine, if I’m trading, say 70 markets with my portfolio for 30 years, that’s 70 times 30 years of possible history which my models are then determining the optimal stop loss.

So that’s how I’m avoiding it being curve fit, because it’s applied over such an extensive data sample that the chances of it being curve fit, for any specific market or any specific condition, is very limited. That’s the way I overcome the overfit side of my testing.

What I’m finding is that I’m also trading an ensemble of trend following models ranging from short to medium to long-term models. That, therefore, also changes the stop loss multiple that’s being applied.

All of my models are applying a consistent 25-period ATR, but the multiples that are applied to those determining the initial stops might be ranging say from 3 ATR up to 8 ATR, but it’s dependent on the short-term, the medium-term, or the long-term model where they’re fitting. The longer-term models obviously have a wider stop, maybe 6 to 8 ATR, the medium-term models might have something from 4 to 6 ATR, and the short-term models might have 2 to 4 ATR. That’s how I’m doing it.

So, when you talk about loose pants, that’s a principle where we don’t want our models to be too prescriptive to attacking a particular form of trend. We want them to be able to capture a large variety of different trends, and particularly these material trends that are often very volatile in nature. So, you do need this freedom for price to move within constraints.

You can’t be too loose pants because you don’t want to find that, after holding your position for six months, it goes back to your initial stop. So, I’ve got a trailing stop that also applies this principle of loose pants. But progressively over the course of time, that trailing stop is slowly ratcheting up behind the price so that when the trend inevitably turns, not too much profit is taken off the table.

That’s obviously also dependent on the long-term, the medium-term, the short-term model. So, when I’m applying this as an ensemble, if you could imagine, each one of my models has fairly loose pants, but collectively, as an ensemble, they have extremely loose pants because they can capture a vast different array of different forms of trend with that ensemble of models working. Some will be whipsawed out, others will stay in.

That’s how I’m trying to get the best of both worlds: having this massive loose pants with my ensemble and giving this freedom to move but also not leaving too much profit on the table when those trends turn.

[00:13:37] Niels

Just out of curiosity or something, I don’t know myself. Do you apply different lookbacks for the same type of model or do you actually say, no, this particular model only trades this timeframe and therefore this stop rule, but it’s the collection of many different models with just one single time parameter that makes up the ensemble?

[00:14:02] Richard

I suppose, Niels, I’ve got about five different core models, one being a Donchian breakout model, another being a moving average crossover model, another being a regression trader model, and Darvas box congestion breakout model. There are five models I’m trading, but within that, I might be trading up to an ensemble of 10.

So, some of the models, like the Donchian breakout, might have two models, one being a medium-term, one being a long-term. It all depends on when I undertake this testing on a yearly basis. As you know, when new data comes in each year, I’m undertaking this process continually.

Sometimes I’ll find that the models I trade, the 10 models I trade for the next year, might have three Donchian breakouts, two moving average crossovers, one regression line trader, one Donchian. It’s how it all comes out in the wash, effectively, through the testing process of my workflow.

[00:14:59] Niels

I, actually, don’t think I’ve come across anyone who trades the Darvas box breakout. Which I think is quite funny, because people may not know that Darvas, I think he was a show dancer, wasn’t he?

[00:15:10] Richard

I think he was. It’s an interesting one, because a lot of people say, well, that’s not really trend following, and in a way, I tend to agree with them. He is more of a momentum trader. But the Darvas box one, actually, leads me to some of these short-term models which I’m looking at integrating into my portfolio. But look, I like to say that I’m always trend following, so I’ll always argue with someone when they say my Darvas box is a trend following model.

[00:15:39] Niels

We’ll keep it between ourselves, Rich. No worries.

The second question that Adam had was, “What would you say is a reasonable sample size in real trade observations to compare against backtest statistics to ensure the strategy is performing as expected? I.E. would it be reasonable to compare say 100 out of sample trades against the backtest statistics or does this number need to be significantly higher? Once again, thank you for producing such high-quality content.” Well, thank you, Adam, for that.

[00:16:10] Richard

Look, personally, for me, no size is long enough as far as a backtest for me. So, the way I approach these markets, and it’s going to come up in the topics we discuss today, Niels, is that I recognize that these markets are these complex adaptive systems. So, over the course of time, they adapt and change. They’re non-stationary in nature, and the future inevitably is going to be different to the past.

So, what I’m doing using my backtest for is robustness testing, effectively. It’s not to have any expectations about what the models are going to deliver in the future. But what it is, it’s an environment in which I can stress test and backtest models for robustness. The way I approach sample sizes, I prefer to consider it not in terms of sample size, but I prefer to consider it in how many different regimes has your model coped over.

The reason I say that is that when you get into the world of high-frequency trading, you can be in a single regime for a period of two weeks and have a sample size considerably more than what I’ve got over my lifetime. So, what I am concerned with is ensuring that I’ve got as much history and as much different regimes as possible to stress-test models.

That means to me, I will always be looking for as much data, as much ways I can stress-test my models as possible. So, I never get this sense of certainty about that’s a sufficient sample size for my models because I’ve always got in the back of my mind this understanding of uncertainty and that it’s best to test it across anything you possibly can in terms of liquid financial markets.

That’s why I test it across every single one of my markets as well. Because if I do have 70 markets and 30 years of history, the way you could look at that is that’s 30 years times 70. That’s giving 70 times 30 possible years of history of different forms of market regime, different forms of conditions to test my models on. So, I’m a big one for huge amounts of data, as much as I can get my hands on.

[00:18:24] Niels

All right. With that out of the way, Rich, it’s time to dive into some of your topics. Today, we’re going to unravel the mystery of complex systems and why regular math just doesn’t cut it. So, we’re diving into complex adaptive systems, or CAS for short. These are tricky beasts you find everywhere from nature to the stock market. They’re super interesting, but a real headache for the usual number-crunching methods.

We, especially, are going to look at financial markets as a type of CAS. Here’s the twist. These markets are always changing and adapting. That means they throw curveballs that standard stats just don’t catch, hiding risks we might not see at first glance. So, with that very brief introduction, Rich, I’m going to leave it to you and hopefully keep up with you.

[00:19:20] Richard

Okay, Niels. Here we go. What I’m going to do, Niels, is I’m going to explain these key properties of complex adaptive systems. These key properties really are game-changers in the way we understand systems and in the way we need to analyze systems. So, if we can appreciate that these financial markets have these things called complex adaptive systems, with these particular properties, it’s going to possibly open our eyes to new ways of analyzing these markets as opposed to the traditional approach that we’re often familiar with and what we often hear in our trading world.

I’m going to describe, firstly, the five key properties of complex adaptive systems and then give examples from both nature and from the financial markets so we can come to grips with what this means. Once we understand those five properties of the complex adaptive system, then we’re going to look at how traditional models might lead us into traps, if we are using those traditional models as a way to assess these financial markets.

I’ll start off talking about these five key properties of complex adaptive systems. The first one is nonlinearity. Nonlinearity in a complex adaptive system means small inputs can lead to disproportionately large outputs and vice versa, making their behavior non-predictable and nonlinear. So nonlinear is his fundamental property found in complex adaptive systems where the relationship between outputs and inputs are not proportional, often leading to these unpredictable and disproportionate outcomes. Nonlinearity, typically, is a result of one-to-many relationships as opposed to one-to-one relationships that exist in complex adaptive systems and can lead to the whole being greater than the sum of its parts.

Let’s look at, firstly, what these one-to-many relationships mean. I’ll give you an example. Think of a rainforest where we have a canopy tree in that rainforest. The canopy tree is used by many living systems for a range of different functions. It provides a place or habitat for certain species in the canopy tree. It provides a food source for many other species. It provides microclimate regulation, significantly influencing the rainforest microclimate. It provides nutrient recycling, water cycle contributions, carbon sequestration, support for climbing plants and epiphytes to grow on that canopy tree, and pollination and seed dispersal.

So, there are many, many different functions assigned to that one canopy tree. If we were simply dealing with a one-to-one function, we wouldn’t see these nonlinear relationships. But when we consider all of these different functions, we can see that we might have one-to-many relationships that lead to a complex adaptive system being more than just the sum of its parts.

So, the multifaceted functions contribute to a web of nonlinear relationships and interdependencies leading to the creation of these emergent properties arising from the single canopy tree. When we consider the canopy tree and the rain forest, now we can look at other complex adaptive systems such as a human being, an elephant, a firm, a city, a financial market, or even an artwork.

These are all examples of emergent systems whose form and function cannot be fully understood by merely analyzing the individual properties of the atoms comprising them. So, at the heart of these emergent systems is this one-to-many relationship, and these plural relationships make these complex adaptive systems nonlinear in nature, which contributes to their complexity and unpredictable nature.

I’ll give you an example of another nonlinear system in nature. This is where we get algal blooms in aquatic ecosystems, you might have heard of, Niels. Algal blooms occur when there’s a rapid increase in the population of algae in water systems. This is often fueled by excess nutrients like nitrogen and phosphorus being added to the water system from sources such as agricultural runoff or sewage.

At first glance, the addition of these nutrients, small as they might be, might seem to be only a minor change. However, this small input can lead to a disproportionately large and often harmful output for that aquatic system. So, this small increase in nutrient levels leads to a massive growth in algae, which is the output.

This then affects various components of the ecosystem, from water quality to the health of the aquatic species, and the overall impact on the ecosystem is far greater and more complex than that initial input of those few nutrients going into that system. This showcases how nonlinear dynamics can lead to unpredictable and substantial changes in these complex adaptive systems.

I’ve given you a couple of examples in natural systems. Now let’s look at examples in the financial market.

[00:25:03] Niels

Maybe, actually, just to interject here, Rich. Actually, in high school, I wrote a thesis in ecology. But I have to say I never expected that was going to come in handy later on in life when listening to you explaining complex adaptive systems.

[00:25:17] Richard

There you go, Niels. It all links together. This is the way trend following works that links everything together. But look, I’ll give you an example of nonlinearity in the financial markets. This is the impact of the social media influencer on the stock market. Imagine a scenario in a stock market, which is a typical complex adaptive system, where a relatively small event triggers a disproportionately large reaction.

Here’s the example I’ll give you. A well-known financial influencer posts a mildly positive tweet about a relatively unknown tech company. Now, the environment they’re dealing with is a complex adaptive system because it includes stock market participants and all of the social media participants that are linked to that stock market with millions of interconnected participants and variables in that complex adaptive system.

So, the initial reaction is that a few of their followers buy stock in that company based on that social media influencer’s point he made in his Twitter feed or whatever. Then we get amplification as these purchases, by those small number of followers, increase the stock’s activity, and it begins to appear on the radars of algorithms and day traders that there’s suddenly an increase in activity; buying activity of this stock.

Then we get this snowball effect where the increased trading volume and price movement attract more attention. News outlets start reporting on the hot stock, further amplifying interest. This leads to this disproportionate response. Within hours or days, the stock price skyrockets far beyond what would be expected from that single social media post.

So, this is an illustration of nonlinearity where the input is small compared to the vastness of the stock market, for example, yet it led to a massive output, significant stock price increase. It’s also unpredictable because the scale of the response was not proportional to the input, making it highly unpredictable. But in a linear system, a small input would result in a small output.

We now understand what this nonlinearity is. Which is that you find that prevalent in all complex adaptive systems. So, we’ll move on to the next one.

These are these emergent properties. This is another fundamental feature of all complex adaptive systems, emergent properties. These are characteristics that emerge when the system components interact in specific ways.

These emergent properties are not found in the individual agents themselves but are found in the way these agents interact. It’s a manifestation of the collective properties of a system, not the individuals themselves. For example, the collective behavior of an ant colony, which seems to exhibit intelligence as a colony, which is an emergent property not found in individual ants themselves.

We’ll look at that ant colony so we can drill down into what these emergent properties are. If we look at each ant in a colony, it operates based on relatively simple rules and behaviors driven primarily by these basic instincts and pheromones. Individually, ants display simple behavior like foraging for food, carrying materials, or tending to the Queen and the larvae. But their collective behavior is different. When these individual ants interact and work together, the colony, as a whole, exhibits behaviors that seem to reflect a higher level of intelligence and complex problem-solving abilities, which no single ant possesses.

We’ll have a look at some of these emergent properties. We’ve got complex nest construction. Ant colonies build elaborate nests with specialized chambers for different functions despite no single ant having the blueprint for the entire structure.

We’ve got efficient foraging strategies where ants find the shortest paths to food sources, collectively, and communicate this information to other ants, optimizing the colony’s foraging efficiency as a whole. We’ve also got adaptive responses to threats. The colony collectively responds to threats, or changes in the environment, like relocating the nest or defending against predators in a coordinated manner. All of these properties are emergent properties.

The intelligence of the ant colony is this emergent property. It emerges from the interactions and communications among the individual ants themselves. It also has a lack of central control. So, there’s no central control, or leadership, directing these complex behaviors of the colony. Rather, they arise from the local interactions themselves and the simple rules followed by each ant.

This example illustrates how emergent properties are a hallmark of a complex adaptive system and they reveal that the whole can, indeed, be greater and qualitatively different than the sum of its parts. In the case of ant colonies, the collective intelligence and problem-solving abilities are emergent properties that arise from networks of interactions among the individual ants. So, that’s now looking at it from an example of nature.

Now let’s look at emergent properties in the financial markets. The concept of emergent properties in financial markets are where interactions between, once again, numerous individual agents. Instead of ants, we’ve now got traders and investors. We’ve also got institutions. These interactions lead to emergent behaviors that cannot be attributed to any single participant themselves.

Here’s how the analogy plays out. Each market participant, whether an individual investor, a large institutional player, or an automated trading system, operates based on their own strategies, goals, and information. These strategies may range from simple buy-and-hold approaches to say complex algorithmic trading, for example. Now let’s look at this as a collective.

When these individual agents interact within the market ecosystem, they collectively create market phenomena that no single participant could generate alone. Here are some examples of emergent properties. Market trends and movements. These overall market trends, bull and bear markets, are emergent properties. They result from collective buying and selling decisions of all market participants influenced by a multitude of factors, economic data, company performance, global events, market sentiment, etc.

Another emergent property is price discovery. The process of price discovery where the market price of an asset is determined is an emergent property. It’s the outcome of numerous buy and sell orders interacting in the market. We’ve also got market sentiment, another emergent property. Collective emotions like fear or greed can drive market rallies or crashes, and these are emergent properties. They arise from collective psyche of market participants reacting to news, rumors, market conditions, etc.

We’ve also got market resilience and crashes. Resilience in financial markets or sudden market crashes, like 2008, are emergent phenomena. These phenomena result from complex interactions among various market players, policies, economic conditions, and so on and so forth.

So, just like the collective intelligence of an ant colony, the trends, prices, sentiments, stability, in financial markets are all emergent properties and they’re not directed by a single entity, but arise from the intricate web of interactions and transactions among diverse market participants.

This creates complexity and unpredictability because the complexity leads to the unpredictable nature of the markets, like the emergent behavior in the ant colony. A market’s response to a new piece of information or change in policy can be disproportionate, can be unpredictable, reflecting a complex interplay of these individual actions and reactions.

So, that’s the emergent property. That’s the second major feature of complex adaptive systems. We’ve seen this in ant colonies and we’ve seen this in the financial markets.

The third very important property is self-organization. A complex adaptive system has the ability to spontaneously create order. A classic example is the flocking behavior of birds in nature where complex patterns emerge, say with the starlings, without any central coordination. This is an example of self-organization.

We’ll look at this flocking phenomenon. When birds flock they form highly coordinated and fluid movement patterns which we’ve often seen, which are these mesmerizing, undulating clouds of movement by the birds. You’ve seen them, Niels. I’ve seen them as well. These rules, that the birds are applying, are based on simple local interactions. Each bird simply responds to the movement of its immediate neighbor without any awareness of the overall shape or direction of the entire flock.

They’re looking at these local rules at the individual level. So, this emergent flocking behavior is a collective phenomenon where we get cohesive alignment. The birds avoid obstacles as an entire unit. But it’s not that they have observed those obstacles, it’s just that they’re obeying these local interactions with their nearest neighbors that allows the entire flock to avoid the obstacle.

These principles of self-organization, when we apply it to the financial markets, here’s a couple of good examples when we get to the financial markets. Individual decision-making occurs by each individual market participant where they’re making decisions based on their personal strategies, information, goals, etc. Then we get these local interactions where these decisions are influenced by, and influence, the decisions of others locally; their neighbors, their friends, their family, etc. Much like the birds responding to their neighbors.

Traders respond to price movements, news and other traders’ actions. Then we get this emergent market behavior; tje overall trends, market sentiments, movements in the financial markets, they emerge from these myriad of different local interactions. But no single participant or group controls these movements, much like the single bird doesn’t direct the entire flock. So, that’s an example of this self-organization that we see in complex adaptive systems.

The fourth major property is this function of adaption and evolution, hence why they’re called complex adaptive systems. Complex adaptive systems can learn from their environment and evolve over time. The financial markets are constantly adapting to new information. Here is a really good example, in the financial markets, where financial markets adapt and evolve.

In the early stages of electronic trading, financial markets were dominated by traditional trading strategies where decisions were often made on slower timescales based on fundamental analysis of companies and markets. We can go back to the 1980s, 1970s. This was the way the markets operated back in those times. But then we got an introduction of new technology, where the advent of advanced computing and communications technologies. A new form of trading emerged, high-frequency trading using complex algorithms to execute trades at incredibly high speeds, often in fractions of a second.

So, we have the market adapting and evolving to this new player, this new process, this new behavior now that has emerged into the market. This is where the system has learned from the environment. As financial markets evolved with these electronic trading platforms, high-frequency trading algorithms began to learn and adapt to market conditions, leveraging speed and efficiency to gain an advantage over traditional trading methods.

The market’s response to this was, well, the markets adapted to the process of high-frequency trading. They included changes in market structure, the introduction of new order types and trading rules and adjustments in strategies by traditional traders who now had to account for the impact of the high-frequency traders on market dynamics.

We had the evolution of strategies such as overtime, HF too became more prevalent. Its strategies has also evolved. This included developments in machine learning, artificial intelligence to analyze market data and execute trades based on complex adaptive algorithms. We got a continuous evolution in markets. New regulations came into play into the markets to respond to the impact of high-frequency trading in the market. We had further technological advancements in the development of counter strategies by other market participants.

These are all examples of how complex adaptive systems adapt and evolve overtime. They can be such significant changes in the market that we never get, actually, a repeat of history. We get a new regime, a totally new arena, a newer environment.

When you look at things, in terms of complex adaptive systems, and when we look at things such as the evolution of natural systems, we see this new environment allows new opportunities to flourish, new species to arise. Traditional species get vanquished or outcompeted by these new emerging species.

We can see that this process is this continuous adaption and evolution. But it doesn’t occur in a gradual way. It often occurs in this punctuated way where we get a period of stability followed by a major change, evolutionary change to that environment, a new environment evolves, and then we get an explosion of new species, new behaviors, new opportunities. That’s the way complex adaptive systems evolve into the future. This has significant impacts on the way we’ve got to consider how we should trade these markets or consider how we should analyze these markets.

The last major property of a complex adaptive system are these feedback loops. These feedback loops can either be positive or negative, and they either amplify or they dampen changes within the system. Feedback loops play a crucial role in the dynamics of these complex adaptive systems, including financial markets. These loops can either amplify positive feedback or dampen negative feedback changes within that system leading to a variety of different outcomes.

Here’s an example of a positive feedback loop in the financial markets. This is where we get a rising stock market, for example. Initially, we get an increase in stock prices. Suppose a particular stock or sector starts to perform well due to strong fundamentals or positive news. We then get investor enthusiasm. This performance attracts more investors, driving up demand and consequently prices. Then we get media attention and hype coming into the market.

As the prices rise, media coverage increases, drawing even more attention and investment to the stock. As we’re going through these processes, this is perpetually amplifying the move, amplifying the move. Then we get this continued price rise with a cycle of rising prices attracts more investors, which in turn drives prices even higher in this positive feedback loop. It can lead to significant overvaluation bubbles, those sorts of aspects.

But then we can also have negative feedback loops in the market, such as we experience in corrections where, at some point, the market may start to recognize that stock prices are overvalued. The selling begins. We’ve got investors beginning to sell their holdings to realize profits. We get a price drop, and then more people sell. The prices start to fall, which triggers further selling.

Ultimately, we get stabilization where the selling and price decline is a negative feedback loop, and this actually helps the market to self-correct, bringing stock prices back to more reasonable levels. This amplification and dampening are the results of this positive and negative feedback loop, and this is a critical fundamental property of these complex adaptive systems. So, they’re the five points, the key properties.

Before I get into the next phase, Niels, which is now saying, what are the pitfalls in the way traditional statistics might be overlooking some of these important impacts found in complex adaptive systems and what could we do to perhaps other ways to look at how we can accept that these properties exist and perhaps better manage risk under uncertainty, under these adaptive conditions, that sort of thing? So, I’ll go into that in the next phase. Are you still with me, Niels?

[00:43:13] Niels

I’m still with you. Of course, in the interest of time here, we will probably only touch on a few things. I think getting the fundamental understanding of these five properties are very important, even though it can feel a little bit heavy listening to it.

But one thing, and maybe we can just talk around that at the same time, is that we talked about all these things that can influence the markets, all these different properties that can influence.

And yet in some ways, when we talk about our field of investing (so trend following), we kind of talk about the fact that we tend to do the same. Meaning, yes, we make tweaks, etc., but, to a large extent, we are certainly applying the same “golden rules” that we have done for decades. Sure, we improve and sure, we have differences in terms of how we manage to trade and so on and so forth.

But, to a large extent, I think our argument has been that our approach has been able to cope with many different regimes, which is a result of a lot of these impacts that you talked about. So perhaps you can weave that in a little bit, and maybe just on a high level talk a little bit about some of these things that can deal with, or better deal with, the shortfalls of traditional statistics in this case.

[00:44:47] Richard

You raise a really good point, Niels, because this is where, I think, trend following models tend to shine or outclass alternative investment processes. Because trend following models, at least to my understanding, are based on the principle of a model that can survive over many different forms of different regimes. The process we use to test robustness for our models is of vast backtests over many different regimes.

We’ve got these very simple models which are not over-optimized for any specific market condition. We’re not looking at finding the most profitable trade in a particular period of time based on a pattern. We’re looking at something that can survive but can exploit these unpredictable opportunities that we can’t predict in advance. This is where our models differ from other models which are targeting a particular edge based on a recurring pattern or feature that they’ve identified, maybe through pattern recognition or whatever.

Our process is more aligned to surviving robustly and being in a position to capitalize on opportunity. They’re much more adaptable in their abilities to adapt to new regimes than other models that are more optimized for a particular class of pattern or a particular property of those financial markets. This is where I think that trend following models do shine because of their ability to withstand the elements effectively, and also, when these opportunities arise, exploit and capitalize on those opportunities.

[00:46:30] Niels

I’m not sure how to phrase this elegantly but when I think about trend following and when we talk about it as well, and this comes from the way I think we think about investing, and that is we say that we are first and foremost risk managers. We then go on to say that we don’t control the performance, we’ll take what the market gives us, but we do have some influence on the risk. That’s why we say we’re first and foremost risk managers.

Do you think that maybe that simple but very important starting point, so to speak, compared with maybe other strategies that maybe look at the return potential before they look at the risk, I don’t know, is what makes the strategy so robust, and, as we know, where we have track records that goes back five decades and have survived incredibly different market regimes?

[00:47:36] Richard

Yeah, I totally agree with that. Because the way I see our models, when we come down to the golden rules, inevitably, all of us trend followers, we like cutting our losses short. We do it in different ways, but we all prevent our losses from being adverse left tail events. So we’re always stunting those losses to as small as possible. We are leaving ourselves open for unbridled opportunity.

When you look at the nature of these complex adaptive systems, what I’ve been talking about today, we notice how unpredictable these markets are; how dependent on nonlinear dynamics they are. These are the properties that give these significant opportunities in these financial markets which these traditional models are very poor at navigating.

But our particular technique, the golden rules, we’ve put it into the furnace at Mount Doom of data testing over huge amounts of market regimes, we know that they stand up in these sensitivity tests, these robustness tests. They stand up to all of what history can present in the past, plus more if we’ve got other different tests we undertake to test for robustness.

We know that they’re going to be the cockroach models. The models that survive into the future, where other species that are trying to target a particular opportunity or exploit an opportunity. We’re a bit more suspicious about when and where these opportunities occur. We don’t like to predict in advance. We just know that markets have this unbelievable ability to deliver this simply because we understand that there are these complex adaptive systems which have these amazing properties that we’re not looking at in perhaps the right way.

That’s why I’ve got so much faith in our models. It’s a bit like the most robust car or the car fleet. That’s what I think I’m trading when I’m trading my trend following models.

[00:49:40] Niels

Do you think you can just maybe very briefly touch, and maybe we’ll keep the more detailed stuff for a later conversation, Rich.

[00:49:50] Richard

I’ll break it down as dot points effectively, so I’ll really get through this next part. We’re looking at some of the limitations of traditional statistics based on these key properties we’ve discussed in this podcast as five key properties of complex adaptive systems.

Now, the first thing is that a lot of traditional statistics deal with linear statistics; normal distributions, we’re dealing with Sharpe ratio, that’s a linear statistic. Standard deviation, that’s a linear statistic. Correlation measures is a linear statistic.

I’ve talked at length, earlier, about nonlinearity in complex adaptive systems. So, these linear models aren’t necessarily capturing that nonlinearity that is expressed in so many complex adaptive systems and these financial markets. In a nonlinear system, the relationship between variables is, really, a simple linear correlation.

Financial markets are influenced by a multitude of interconnected factors. Economic indicators, political events, investor sentiment, these factors interact in complex ways. A linear correlation is not going to cut it as far as identifying if that relationship is going to persist into the future because we’ve got this dynamic interplay of all of these interwoven events, that’s going to shift that correlation around making it very non-stationary in nature. So, all of these linear measures are effectively assuming that the market is stationary.

We know that markets are non-stationary in nature, and they can do weird things that we haven’t experienced in the historical track record. That’s about linearity. We want to avoid using statistics that… Well, if we do use these linear statistics, we’ve got to be aware of the shortfalls of them in dealing in a nonlinear environment.

The second thing is traditional statistics often deal with what we call reductionary processes. What happens is traditional quantum methods try and break down systems into simple isolated parts. In this process of analyzing things in these simple, isolated parts, they miss the whole interrelationship between agents, between system elements that are so important in complex adaptive systems. This, therefore, often means that they totally underestimate risk events, because they don’t see the interrelationships and dependencies that exist in that complex adaptive system.

We see that analyzing market sectors independently in financial markets is not necessarily a wise practice. A common method in financial analysis is to examine individual market sectors like technology sectors or healthcare sectors in isolation to assess their performance and risk. But the limitations, when we look at it from a complex adaptive system, is that this approach overlooks how sectors influence each other.

For instance, a technological breakthrough in the renewable energy sector might significantly impact the oil and gas sector, and vice versa. We’ve also got the single factor models that are applied in economics. There are significant limitations to that. So, this is part of this reductionary approach. In economics, these models often focus on single factors like interest rates alone, and predict outcomes like investment or consumption based on these single factors.

But the limitations, in a complex adaptive system, are that these economic systems are influenced by many interconnected factors: political events, technological changes, social trends. These factors interact in very complex ways, and single factor models simply can’t capture that.

The other example of the reductionary reprocess at play is where people are looking at things like individual trading behavior in stock markets. So, a reductionary approach would analyze individual trading behaviors to understand market trends. They wouldn’t look at how a complex adaptive system views it where the stock market’s movement is a result of collective behavior, where individual actions are influenced by and influence others in the feedback loop. So, that’s the second thing we’ve got to be aware of are these traditional models, this reductionary process; avoid the process of reduction and analysis because you often miss the important connections and interrelationships that exist in that system.

The third problem of the traditional methods is related to predictability and determinism. Complex adaptive systems are inherently unpredictable due to their evolving nature, but traditional models, which often seek deterministic solutions, are ill-equipped to handle this unpredictability. An example is a dependence on backtests to estimate future returns.

The traditional approach uses backtests as a basis to derive an estimate of expected returns in the future, or measure risks, through variations from expected returns. But the challenge in a complex adaptive system is that these systems evolve and adapt over time and the past can be an unreliable guide to the future. New emergent properties arise, which significantly reshape trajectories. As a result, we see new risk events we’ve never observed in the historical record.

So, there are many shortcomings when you start looking at the dependence of backtests in your analytical process. Strategies often are overfit for the past and very underfit for the future. They carry warehouse risk attributed to events that have never occurred in the past. But they’re likely to occur in the future because of the adaptive evolutionary nature of the markets.

We’ve got things such as mean reversion trading strategies. Mean reversion trading strategies, they’re based on the assumption that the price of an asset will tend to revert to its historical average over time. There’s a challenge in complex adaptive systems because we know financial markets can experience prolonged periods of trending behavior that significantly deviate from these historical averages. This creates these risk events for mean reverting systems.

Also, we’ve got a problem with econometric models for market prediction. The traditional approach is that economists often use historical economic data to predict market movements, assuming stable relationships between economic indicators and market performance. This is a problem in a complex adaptive system because the relationships between these economic indicators and market performance changes over time due to a variety of factors, policy changes, technological innovation, and it just goes on and on.

The fourth major problem, associated with the traditional way of looking at these markets, is it overlooks these emergent properties that we’re talking about. By focusing on individual components, traditional methods often miss behaviors that are crucial to understanding a complex adaptive system.

Emerging properties relate to these complex behaviors and patterns that arise from the interactions of multiple market participants, etc. We’ve got market sentiment and herding behavior. These are all emergent behaviors which need to be taken seriously by traditional economic models.

We’ve got systemic risk. Traditional analysis focuses on risk profiles of individual financial institutions or instruments. But in complex adaptive systems, which have emergent properties, systemic risk often emerges from a network of interconnections in the financial system. The failure of a specific broker can cascade across the entire financial system because of these interrelationships and interconnectedness. Now, we get the major broker collapse, or the Lehman Brothers collapse, and it infects the entire financial system.

The last point, Niels, before I close up on this is this historical dependency, because complex adaptive systems, they all are influenced by their past states, a factor that many traditional models do not adequately account for. This is this process we talked about previously, Niels, called serial correlation. Dependent on the sequence of events, when they arise and what type of events they are, they do shape future events.

So, there is this historical connection. We can’t just assume that there is no significance in past events. They carry a momentum signature. They carry signatures that do, actually, go into the future to create these unusual future trajectories. For example, consider the impact of long-term economic cycles such as those described in your book “The Fourth Turning”, Niels, in the fourth turning where we have Western capitalist economies experiencing long waves of booms and busts lasting between 50 to 60 years between these events.

These are all these historical dependencies that exist in the market that, actually, influence future states of the market. We’ve got path dependence in asset prices. We don’t need to go into that. We’ve talked about that enough on this episode, Niels. Impact of past financial crises, so the influence of crises like the 2008 global financial crisis on subsequent regulatory changes, investor behavior, market structure, these all have influences into the future.

So, Niels, the problem with these traditional models is that they’re not addressing some of these key features of these complex adaptive systems. But there are particular processes we could be looking at and, perhaps, should be more strongly looked at. Not to go into to much detail, but agent-based models are a really interesting area that, when undertaken correctly, can take heed of all of these properties we’ve seen in this complex adaptive system where we use these agent-based models where the agents consist of various types of trading agents. You do this in a computerized environment where you include different populations of different behavior of traders who base their decisions on different aspects such as fundamental value. We have noise traders in the mix who make these random, uninformed decisions. We’ve got technical traders who are relying on historical patterns. When we put these collections of participants into these agent-based models, we get these relationships and independencies all developing in these computerized simulations which start showcasing these periods of boom and bust that aren’t being picked up by these traditional models. It’s showcasing a lot of the symptoms we’ve been talking about here.

So, it’s a really exciting area, these agent-based models, and that’s possibly where the future of understanding these markets is. There might be more emphasis placed on this in the future rather than the traditional focus on say the linear models and those sorts of things. But look, I’ll leave it there, Niels. We’ve explored a lot in this episode. I probably haven’t been able to do justice to it totally, but I think I’ve got across the main points here. Any thoughts?

[01:01:18] Niels

This stuff is complex. I think, even for people who have been in the trend following world for a long time, some of the things that you’re bringing to light, I think, are looking at this from a different perspective, explaining it with different words, different terminologies than maybe most of us have been doing when we’re thinking “a simple” strategy such as trend following.

So, it’s super important, but it can be very difficult to absorb. Therefore, it’s a conversation that needs to continue, so to speak. I really appreciate you helping all of us to appreciate the challenges, yet also that there are solutions and solutions that, in a sense, we as an industry have been embracing maybe without thinking about it in the terms that you explain.

How you link it back to nature makes it, obviously, understandable that these forces are probably never going to go away, which is the whole reason why trend following continues to work. It is based on, well, I say human behavior, even though we’ve talked about the rainforest and ants, which is not quite humans, but it is incredibly important to understand that.

[01:02:44] Richard

What I find interesting is that if we go back to Katy Kaminski’s podcast, when she was talking about the properties of trend following models, their adaptiveness, their flexibility, those points that she raised, if you refer to those points, you’ll see that those points are fully embedded within these complex adaptive systems.

I often feel that trend following, out of all of the strategies I know about, and I know quite a few of them, this is the one most aligned to what we’re talking about in a way to navigate this uncertainty in these complex adaptive systems.

[01:03:17] Niels

Absolutely. I completely agree. I think most of the people following these conversations and listening to us, they will, for the most part, either have experienced or at least heard of the great financial crisis and how it just felt very different from many other crises. Because you suddenly were seeing banks go down and what effect did that have on other banks and so on and so forth, and then the whole financial system became very fragile.

So, it’s certainly, for me maybe, the easiest way to think about this complexity and how changes in one part of the system can have huge effects on other parts of the system. It goes back to what I often say, which is not a very elegant way of putting it, but it’s this imagining the unimaginable.

This is what we need to build into our investment strategies are strategies that can, essentially, cope with the unimaginable. Every time we have a crisis, or you could say in a day-to-day sense, there’s always something new that we haven’t seen before. That’s obviously what makes investing super challenging, but also where these rules come into play and why they’re so critical to long-term success, at least.

Those are my thoughts, Rich. As I said, much appreciated.

Now, in a couple of weeks, we will be releasing our annual year-end conversations. We’re going to do that. They’re going to be the last two weekend episodes of the year. I would appreciate it if our listeners have any topics that they would suggest that we dive into something that we… We’re going to be a lot on the recordings.

We’re going to be, all of the co-hosts, coming on. So, there should be something for everyone. But of course, it’s important to find some topics that are of interest to the listeners. So if you want to send me at info@toptradersunplugged.com, something that you think many of us could chime in on that you find really important at this point in time, ideally related to trend following or volatility or something like that, we’ll certainly do our very best to bring it up. But of course, I can’t promise that all of the topics that we receive will be discussed, but we’ll do our very best.

Of course, we would appreciate it if you would share and rate and review these episodes because it is so helpful to us in order to continue to grow the show. Next week, I’m joined by Nick Baltas, from Goldman Sachs, so that will be your chance to get him to answer some of your questions. Feel free to email them to the usual email address info@toptradersunplugged.com.

Rich, thank you ever so much for this masterclass in complex adaptive systems. This is definitely your home field, and I look forward to continuing to explore this and better appreciate, myself, some of these really fascinating aspects of the market. So, thanks again, Rich, for doing that. Until next time, take care of yourself and take care of each other.

[01:06:53] 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.