In the Beginning there was Trend Following – A Primer – Part 2
Trend Following Primer Series – Care Less about Trend Form and More about the Bias within it- Part 2
Primer Series Contents
- An Introduction- Part 1
- Care Less about Trend Form and More about the Bias within it- Part 2
- Divergence, Convergence and Noise – Part 3
- Revealing Non-Randomness through the Market Distribution of Returns – Part 4
- Characteristics of Complex Adaptive Markets – Part 5
- The Search for Sustainable Trading Models – Part 6
- The Need for an Enduring Edge – Part 7
- Compounding, Path Dependence and Positive Skew – Part 8
- A Risk Adjusted Approach to Maximise Geometric Returns – Part 9
- Diversification is Never Enough…for Trend Followers – Part 10
- Correlation Between Return Streams – Where all the Wiggling Matters – Part 11
- The Pain Arbitrage of Trend Following – Part 12
- Building a Diversified, Systematic, Trend Following Model – Part 13
- A Systematic Workflow Process Applied to Data Mining – Part 14
- Put Your Helmets On, It’s Time to Go Mining – Part 15
- The Robustness Phase – T’is But a Scratch – Part 16
- There is no Permanence, Only Change – Part 17
- Compiling a Sub Portfolio: A First Glimpse of our Creation – Part 18
- The Court Verdict: A Lesson In Hubris – Part 19
- Conclusion: All Things Come to an End, Even Trends – Part 20
Care Less about Trend Form and More about the Bias within it
When we begin our trend following journey it is easy to be drawn to the idea that we need to classify trends and be prescriptive in the form of trend that we wish to trade. Unfortunately, such a narrow focus can mean that we filter out a vast array of potential trending conditions that may have an enduring directional bias impregnated within them.
Furthermore, can we really rely on our brain as the arbiter of truth? Our brains are engineered to find causality in visual patterns. This can lead us to false conclusions when simply using visual clues alone as a basis to infer meaning into trend form. The host of different interpretations surrounding the context of different forms of trending condition requires us to adopt systematic rules-based interpretations using the data itself to convey any gross statement of statistical meaning such as its directional bias and overall volatility.
We therefore need to understand that while tempting, we should not focus our attention overly on trend form itself. There is little to be gleaned for a trending price series that can help us ‘trade the trend’ apart from the overall bias and the volatility.
Fortunately, we do not need to pay that much attention to trend form as when we are trading on the right-hand edge of the chart, the ultimate form a trend takes can only be determined in hindsight. Without a crystal ball, the actual form a trend takes to its conclusion is an academic exercise and does not really matter.
While we say that we are trend followers, what we are trying to achieve is to use trends as simply a method to enter a market with our ultimate destination unknown. By only entering a trade during trending market conditions we are filtering our entry method to avoid random or mean reverting market environments. Also, by only focusing on the most significant trends and catching them late in their evolution, we are basing our expectation on a principle found in liquid markets that the more fully formed trend has a greater probability of containing serial correlation within it. With serial correlation embedded within a trend, then perhaps there is a chance that this trend that we are riding will turn into an exceptional outlier.
The Many Forms of Trend
Trends come in many forms. There are countless definitions of a trend found throughout technical analysis literature from formal definitions requiring a few consecutive higher highs and lower lows, to trends defined by moving averages of a price series and so on and so forth.
While trends appear to be easy to visually spot, when it gets to their specific definition, it is not such an easy task.
In Chart 1 below, we can see 16 highly liquid markets spanning several asset classes such as Forex, Indices, Soft Commodities and Metals charted on the monthly timeframe. Visually the trends appear obvious but is extremely difficult to actually define the ‘trendiness’ of markets in categorical terms that all trend followers would agree on.
Chart 1: Market watch for the Monthly Time-frame
Now while many may say in recent times that ‘The Trend is Dead’, Chart 1 clearly refutes this claim stating that at least from a visual perspective trends are alive and well. So, what gives?
While trends clearly persist right up to the present day, the degree by which trend followers can take advantage of these trending environments has become far less simple. The ‘trends of today’ are not as clear cut as the ‘trends of yester-year’. Trends today are more volatile with considerable variation of form. Market participation has changed, and trading behaviour has become more complex. A myriad of speculators now apply a vast array of trading behaviours which considerably alter the nature of trending condition.
Remember that complex markets are adaptive in nature. Trends are emergent structures of financial markets representative of market disequilibrium (periods of market transition) which cannot be categorized into a single definition. Unlike predictive oscillations of price action, the trends that we want to catch (aka the outlier) are inherently ‘uncertain’ in nature and can be found in the fat tails of the market distribution of returns (more on this later).
As a modern day diversified systematic trend follower, our aim is therefore to be less definitive in our opinion of what constitutes a trend as we may find that such an approach encourages us to miss many trends that ultimately manifest as outliers.
Furthermore, with our approach to diversification, we want our systems to capture a variety of trend forms and segments to provide much needed correlation benefits at the portfolio level. Having a diversified portfolio of trend following systems allows us to attack trends across different markets and timeframes significantly increasing the variety of trends that we can participate in.
The ultimate form that trends can take are spread across a spectrum. They may include:
- parabolic trends that start with an initial humble trajectory and then take flight with an exponentially rising or falling directional move of arbitrary duration and degree;
- to perfectly formed trends that closely approximate a linear regression and are easy to trade to;
- slowly evolving and volatile trends that are particularly hard to trade; and
- any combination thereof.
Now as seasoned trading veterans are aware, the systems we use to capture trending market conditions comprise constraints that dictate the type of trend or trend segment that the system can capture. The rules applied to the entry condition, the initial stop and the trailing stop can be thought of as constraints (like a container) that filter the types of price action that can survive while existing within these constraints. As soon as the constraints are breached, then as per the definitional constraints of that system, the trend is said to end.
While the system rules define what trends can be traded, this does not mean that a trend condition is over. It may elect to resume its prior trending behaviour, or take a breather or often, go counter-trend to the preceding prevailing trend direction for a period of time.
Clearly the system you apply is scaled to a particular form of trend or trend segment whose definition is dictated by the system. So, for a diversified systematic trend follower, they adopt a diversified suite of trend following systems targeted to capture a myriad of forms of different trend.
The Brain likes to assign Causality to Visual Patterns
One of the major issues surrounding trend following is that people like to infer meaning into trend form. This is understandable as our brains have evolved in a natural environment where an edge has been conferred by natural selection through our ability to ‘predict’ outcomes from sensory clues. This ‘precognition’ appears to have given us a distinctive survival edge against those species which have simply ‘reacted’ to environmental threats. The competitive landscape of our natural environment is a hostile one. Any small edge which confers a survival advantage is received by the human species with welcome.
We need to always remember that our brains are encased in our skulls and may not faithfully render the reality that is ‘actually out there’. All it can do is use the initial stream of sensory data inputs to ‘best’ match (aka predict) those inputs against past experience or learning held in memory. It interprets pressure vibrations of molecules as sound and photons received on a retina as sight. If the prediction is right, then our mental models need no adjustment, but once the data stream has been fully received, we can then test the efficacy of that prediction against the actual data stream itself. If there is a predictive error, then the ‘missing’ additional data can be added to our memory for future reference. This process is something we call learning.
Now while this advantage has been extremely useful for our species survival, the predisposition for our brains ‘to predict’ can create problems for interpreting other less familiar complex systems within which we have not evolved.
We must continually remind ourselves that our natural environment and financial markets are different complex adaptive systems. Unlike our natural environment where competition is played out between different species, the financial markets comprise a single species where ‘intelligence’ is par for the course. In such a different environmental context, being a successful predictor does not necessarily convey the same advantage to survival.
Let us have a look at how the predictive nature of our brain may not necessarily convey the same advantage that we have obtained within our natural environment. Look at the results of how our brains interpret the following examples. Now the issues associated with interpreting these examples has been made clear, but just imagine how without applied attention we could easily run into problems if we relied on our brains to interpret market data.
Here are some simple examples to mess with your brain.
Photo 1: Rabbit or Crow?
Photo 2: Troxler Fading (The Lilac Chaser)
Photo 3: Wash your Mouth out with Soap
Having an understanding of the inherent tendency of our brains to predict ‘causality’ into form should create concern when we rely on the brain to visualize chart patterns. The vast amount of price data may or may not possess a causative influence in it, but it is exceedingly unlikely that we will be able infer great meaning from these patterns using our brains.
What we need is an objective way to assess trends which at least we all can agree on. While most trend traders have their own personal ‘definition of trend’ it is not clear whether these definitions have any more meaning than a random plot of market data. It would therefore be very handy if we could apply a principle that standardizes our method of analyzing trending price data. So in the spirit of this agenda, I have had a go at creating my own method of objectively analyzing trends using objective rules based processes that we all can agree on.
Here is how I do it. Using a standard deviation channel, it is quite easy to define the overall bias of the closing price of a time series bar chart by extending the Standard Deviation Channel (SDC) from the lowest lows (swing lows) and highest highs (swing highs) of the series. I can then extend the standard deviation to enclose the maximum adverse volatility excursion of the trending series and provide a context to understand the volatility of the trend. Note that in a downtrend, the adverse volatility is to the long-side and in an uptrend, the adverse volatility is to the downside.
Chart 2: An Objective ‘Rules Based’ Method to Analyse a Trending Price Series
Having such an objective rules-based process to define the directional bias and volatility of a trending price series, we can now revisit Chart 1 and apply this process to observe the significant variation in overall characteristics (refer to Chart 2 above). Using this rules-based process we therefore do not run into the ‘classic’ problems arising from a trend followers’ interpretation of a trend that may be riddled with visual biases.
Chart 3: Market watch for the Monthly Time-frame using a Rules Based Approach for Trend Interpretation
We can now revisit our statements regarding the plethora of various trend definitional forms. We can see from our rules-based application that the bias (slope) and the volatility of trends and trend segments (defined by the width of the SDC) vary considerably across these different markets. Given the plethora of form, there is not a ‘one size fits all’ solution to this riddle.
Having a single definitional approach to trend following significantly hampers our trend following objectives. What we find is that our overly prescriptive approach to what defines a trend significantly reduces our ability to participate in clearly trending market conditions.
So, let us drop the categorisation scheme and go for the simplest statement that catches all. A trend is simply a directionally biased price series that can take a broad number of different forms. Having such a simple definition ensures that our systematic rules-based approach adopts very simple system designs that allow us to have the greatest chance of capturing the greatest variety of trend forms and/or segments.
While I have provided a simple method to objectively define a trending condition, in our world of systematic application, we adopt systematic rules-based methods that avoid the temptation for our brains to impart biases into any form of price action. Each of our trend following systems are used to objectively assess price action without the possibility of imparting any form of discretionary bias. It is through this application of quantitative rules based approaches which allows us to therefore conduct back-tests over market data that are repeatable and free from discretionary judgement.
Do we care about trend form when Trading the Right-Hand Side of the Chart?
Now while we have been discussing the many forms of trend, does this really matter when we are trading the right-hand side of the chart. At the point of entering a trend, all we can see ahead of us is uncertainty. We do not have a crystal ball to tell us the exact form this trend will take. To do so is an exercise in hind-site navel gazing.
You see we do not care about the different form of trends at all. All we really care about is whether the current trending condition we decide to ride may have endurance.
As soon as we jump on board a trending market, the ultimate trending form could take ‘any form’. The reality is that we are not looking for a particular form of trend at all, but rather a trending condition that might have a kernel of serial correlation embedded within it which may drive future directional momentum. It is then up to us to have a system that is sufficiently robust and non-prescriptive to give us breathing room while we ride this ‘uncertain’ trending condition.
When considering what we as trend followers are seeking it is sometimes useful to think in metaphors. So, let us equate trend trading to longboard surfing.
Under our analogy we sit out past the breakers and wait for the swells to arrive before we take the plunge on our boards with the hope that these ‘special’ waves might have some enduring momentum cooked up in them. We are all seeking auto-correlation (aka momentum) in our waves, but we have different designs (aka surfboards) to capture elements of it.
We could have many surfboards in our collection in which we need to choose the conditions in which we use it or we could opt for simple designs that work across a variety of different wave forms.
Now do we focus on surfing the near-shore break or do we swim out past the breakers waiting for the sets of ocean swells to arrive? As experienced surfers we know that it is the swells which are our target as opposed to the normal beach break. Each beach has characteristic wave forms associated with the local conditions and geography of the coastline and they rarely have enduring persistence, but ocean swells have broader features and enduring momentum, and the local terrain has far less influence on their overall form.
While we watch for the waves, we are watching for the ocean swells that may occur in autocorrelated sets of waves. Under this arrangement we can be pretty sure that these wave forms have the necessary enduring momentum to give us a great ride.
Rather than catching all waves we are patient long board riders sitting far outside the normal break waiting for the exceptional before we dip our toes in. We do not know the exact form this potential wave event will take, but we have a buoyant simply designed relatively large board that allow us to ride a variety of possible forms. We replace skill associated with the ‘known’ with a design that accommodates for uncertainty.
You see, maybe a prerequisite for trend trading is to become a surfer. Awesome
Chart 4: Surfing the Trend
It is the Bias in the Time Series that Really Matters
Ok, so having equated trend following to surfing let us get back to business in understanding what we really are seeking to capture in our trend following world.
Despite the variety of different forms which can be confusing, what we are seeking in our trend following pursuit is to trade those trends with serial correlation embedded in them. We can therefore adopt a more valid form of categorisation treatment for trending environments. Under this method of colloquial categorisation, we can classify trends into two broad categories of either:
- Random trends; or
- Real Trends.
The difference between the two is based on whether there is an underlying bias or serial correlation present in the price series where a price earlier on in the price series has an impact on a price point later in the series, and the two price points are defined as being correlated. This underlying bias can be thought of as a causative driver that gives a momentum nudge to the price direction.
So here is an example of a trend constructed from random price data constructed from 10,000 price points.
Chart 5: Plot of a Random Price Series of 10,000 data points
For the trader who views these price behaviours all the time, the price series of Chart 5 would be conferred to have a bias in the price series that has led to this trending price behaviour. In other words, the trader may see this price series as being a ‘real’ trending price series dependent on how they define what a trend is.
The problem however is that this was constructed through random price points and by definition, this trending series will have no trending persistence into the future. This price distribution could be plotted within a Normal distribution of market returns.
Chart 6: Plot of the Same Random Price Series with applied Bias (Serial Correlation)
To continue with this example, we now insert a degree of serial correlation into this random price series to obtain Chart 6. Visually this would also be conferred as a trend.
However, unlike the Trend in Chart 5, the bias has persistence and if it persists will have a ‘biasing’ effect on all future price data. We can see that the impact of this bias on the price series creates a ‘drift’ in the series. This is clearly the trend that we want to trade…but it is not the form of the trend that is important to us….but rather the bias (or the drift) that exists in the underlying price series that has causative power.
On a random price series, the impact of a persistent bias can result in many different forms of trending pattern. We cannot be choosy about the trend we want to change by nature of its form as a myriad of forms are possible with bias impregnated within them.
However, for the trader, it is visually impossible to deduce if the price series has a bias in it. This can only be deduced through hindsight statistical treatment. When trading on the right edge of the chart into the unknown, the ability to distinguish between both forms is impossible. For example, if we compare both charts in hindsight in Chart 7, we can clearly see the differences….yet our ability to distinguish between the two trends by visual clues alone is impossible as the bias is represented by the difference between the price series of Chart 5 and the Price Series of Chart 6 and not the form of each trend. It is the bias itself that illustrates the causality of the correlated price data.
Chart 7: Comparison Between the Random Price Series (Chart 5) and the Random Price Series with Bias (Chart 6)
So that puts us in a dilemma as Trend Traders. Do we trade all trends and run victim to ‘false’ trends?
A way we mitigate against the propensity to be caught by the illusion of a random trend is to only participate in trends when they are well formed and have a greater probability of being non-random in nature. We wait for more extreme trending conditions before we elect to participate in them as we need the price series in our data sample to build to sufficient levels where randomness has a lesser say and any enduring auto-correlation or bias in the series has a greater chance of expressing itself.
Unfortunately, there comes a time, despite our deliberate delay, where we need to participate in all trends that meet our criteria of being sufficiently material. If we do not, then we can be subject to what is referred to as a Type 2 Error. Namely, we miss the opportunity to participate in an outlier.
Given the infrequency of major outlier events, to miss one of these significant windfalls can be the difference between success and failure for us trend followers. We need these major windfalls to pay for all those losses we have incurred along the way. So, we must catch all of them. The way we do this is through diversification and enduring patience waiting for these potential events to unfold.
While there is no guarantee that the trend we ride has any causative substance, as we may just be riding an unusually material random trend, by delaying our trend entry to more extreme levels (on the verge of what we refer later in this Primer series as the “Gaussian Envelope”), then there is a greater chance that we have avoided the randomness (or noise) associated with random price series.
Stay tuned for our next instalment in this Primer Series.
Trade well and prosper
The ATS mob