The Problem we are Trying to Address

As retail traders who are seeking to apply the techniques adopted by the professional Fund Managers in the Diversified Systematic Trend Following niche, we hear many statements made about the selection of universe you apply to trading the trend……..but in these statements there is a degree of confusion created for us retail folk as we don’t know in what context these statements are made.

For example some very successful professional Trend Followers say that “you trade any liquid market the same way both long and short under broad diversification”….while other very successful Trend Followers say that “you treat Equities or Indices differently by virtue of the specific characteristics of this asset class or type of Index construction and therefore adopt a preferential bias towards ‘long only’ momentum/trend following strategies”.

This mixture of different opinion creates a degree of confusion in our retail world where we are confined to a more restricted universe of options that cater to our lesser finite capital…such as the CFD and Forex space. The finite capital restriction places further obstacles on a retail trader….where we actually have to form some selective choice in the universe we ultimately select to trade….but we want to ensure that this selection process is based on a quantitative decision as opposed to a subjective one.

Now as retail traders with a finite pot of available capital we unfortunately are severely limited to the degree of diversification we can achieve to benefit from the ‘edge’ that resides in diversification itself….but in today’s competitive market with a vast array of market maker offerings using CFDs and Forex with micro-lot availability….then it appears that it may be possible (within certain caveats) that it is now possible to at least apply these well worn principles of diversified systematic trend following across different asset classes with a degree of success but with far smaller capital available.

While we appreciate that the universe of the professional FM is a wider one and as a result creates a different context around which the statements of the Professional FM’s are made….our more restricted universal context plus our limited capital available for applying the Trend Following (TF) approach creates a need for a more selective posture to be adopted in determining the universe of markets that we can trade.

It must be noted that using CFDs gives a broad range of exposure to different traditional asset classes such as futures markets, ETFs, Indices and stocks etc. ….but there are notable differences in the structure of these products (namely the ‘costs’ of trading these instruments) in our restricted universe that adds a further layer of complexity to this decision making process.

So this Blog series seeks to break down some of the statements made by the industry to investigate whether or not you need to apply a degree of ‘selection’ in your choice of markets to trade using classic trend following strategies for us poor struggling retail traders seeking a more sustainable path towards the future.

For this series of posts we will use very large data sets across fairly long time horizons and across asset classes widely available to the retail trader. This long time horizon ensures that we include a vast array of different market conditions (favourable and unfavourable to trend following) to detect if any clear bias exists in the data series itself that favours a simple trend following approach.

Furthermore our testing will use a very simple and classic trend following model on the Daily timeframe that has very few parameters and meets the ‘robust’ objectives and fundamental principles of trend following in that the model ‘cuts losses short and let’s profits run. The model is a Donchian Breakout technique that includes the following elements.

Simple Donchian Breakout System 

Entry Setup

  1. A  Donchian channel which is used as a method of breakout entry into a price series with long or short momentum. The look-back period of the Donchian channel acts as a filter to restrict entries based on the highest high or lowest low of the last ‘n’ periods.  A shorter look-back of say 50 days means that more entries are taken over the course of the 20 year time horizon than say a 500 period look-back. It is the look-back itself that defines ‘how hard price has to work’ in reaching the highest high or lowest low. Obviously a longer look-back period means that you catch trends later and also catch far less of them in general than shorter look-back periods. We want to assess whether this delay to entering a trending condition is actually advantageous with the Law of Large numbers to the PL or not, and uncover what this potentially means in our interpretation of whether or not there are better places or not in deciding when to enter a trending condition.
  2. A short term (10 period) Simple Moving Average which is used simply as a basis to ensure than when the Donchian channel is breached by current price, that there is short term momentum backing the price move as opposed to a slow progressive creep in price upwards or downwards. We are not testing the SMA itself as this is just a confirmation measure used for our prime parameter we are testing (namely the Donchian Look-back).
  3. An initial stop used for position sizing purposes which adopts a 20 period Average True Range risking a standard $ amount per trade. This stop method ensures that we apply an equal $risk weighting in terms of our position sizing to each possible return stream to allow for a bias to be visualised relating to the ATR period chosen. The method ensures we normalise the impact of trading different markets and is independent of the different pricing applied for each market.

Exit Condition and Trade Management

  1. The Initial Stop discussed previously is used to limit adverse risk exposure upon trade entry and allows for the principle of ‘cutting losses short’ immediately following entry to be observed in the technique.
  2. A stepped Trailing Stop condition based on a 20 period look-back of ATR which is independent to the Initial Stop and allows for both an assessment for when the Initial Stop ATR and the Trailing Stop ATR are the same….and when the Initial Stop ATR and the Trailing Stop ATR are different.  In instances where we have a 2xATR Initial Stop and a 6xATR Trailing stop…then the trailing stop is initiated when price has moved in our favour by 6xATR. Where we have a 5xATR Initial stop and a 2xATR Trailing stop, then the trailing stop is initiated when price has moved in our favour by 7xATR. Thereafter a 2xATR trailing Stop is maintained. A trailing stop technique is applied to ensure that we cut losses short at all times during the trade management process between trade entry to trade exit but most importantly let profits run at all times. No profit target is used to explicitly exploit this fundamental principle of letting profits run. By examining a range of different ATR trail values, the analysis allows us to examine the impact and any bias that may exist from a tighter or wider trailing stop mechanism applied to this technique.
  3. A move to Break-even determined in ATR which is simply used as a method to determine whether such notions of applying this principle are warranted when our initial stop is large (eg. 6xATR) and our trailing stop is small (eg. 2xATR). Under this scenario we need to wait until price moves in our favour 8xATR before a trailing technique is applied of 2xATR. You can see that a move to breakeven at say 2xATR allows for us to at least benefit from the significant positive price price move by lifting our stop at least to the breakeven point to prevent too much profit being taken off the table by an adverse price move following such a large favourable run.

Now an important consideration to note is the following. Given the simplicity of this system and its wide application to a vast array of different forms of trending condition, this simple system can be considered a proxy for a vast number of different forms of trend following system in which each method “cuts losses short and let’s profits run”.

The limited number of parameters used in this test system allow broad conclusions to be reached about the efficacy of trend following in general to capture trending market conditions. If the model was more prescriptive in nature then such a generalisation could not be reached as the system design complexity would ensure that only a specific class of trend would be captured by the system.

In this specific example, the simple system itself has less impact on the overall performance outcome than the nature of the market condition which is beneficial to simple systems of a trend following nature. In other words….trends come in many shapes and size…and this simple system will capture more of them than a more specific ‘constrained’ system design approach.

Our Testing Universe

For the purposes of this research undertaking, we decided to go as wide as we could looking at those markets made available to the retail trader using Metatrader 4 (MT4) from two major Broker data sources…namely Dukascopy and Avatrade.

The long term high quality data provided by Dukascopy is a frequently used standard data source for retail traders seeking micro-lots in the Forex spot market, a small offering of spot commodity CFDs and also some notable Indices. The data quality is high and the length of data series is long enabling long range testing. 

The market offering by Avatrade using CFDs is significant allowing retail traders using MT4 to gain cross asset exposure using CFDs in micro-lots to a variety of futures commodities, bonds, metals, energies etc. 

So we searched far and wide to obtain as broad a range of different liquid trade-able markets (with micro-lot sizing) and as long a data history as we could get our hands on, to come up with the following broad listing that we could undertake our testing on.  Below are details of the Testing Universe used in this study comprising 63 different markets spanning a number of asset classes.

Spot Forex – 29 Markets

Equity Indices (CFDs) – 11 Markets

CFDs (Other) – 23 Markets

Findings and Conclusions – Part 1

The Bias that Exists in the Entire Universe

Now before we unpack this puzzle, you need to understand the following. Do not make sweeping generalisations based on the results derived from this study applied to the entire universe. We need to unpack it further when assessing the individual asset classes that exist within this restricted retail universe that will help us more definitely judge where the favourable bias exists for trend following that we can exploit. This unpacking will be examined in  future Parts of this Blog series. 

All that we can confidently declare so far as summary conclusions from the analysis across the entire universe are as follows:

  1. The ‘cost drag’ of trading this universe is sufficiently large to overcome any favourable bias that exists in the choice of individual parameters used (in isolation) towards trend following when applied across all instruments in this entire universe. This means we need to find those asset classes where the ‘bias after costs’ are tilted more towards our favour with the Law of Large Numbers.
  2. This study first looks at the ‘favourable bias’ in the values chosen of parameters in isolation, not in combination. For example we isolate the Initial Stop values by ATR (Parameter 1), Trailing Stop values by ATR (Parameter 2), Donchian Period by look-back period (Parameter 3) and BE level by ATR Multiplier used (Parameter 4). Our assessment then looks at these values per Parameter in isolation to see where there is any cluster (bias) towards higher average trade profitability with a particular choice of value used. ….but we recognise that it is actually the combination of how these parameters work together that can turn a losing system into a winning system. 
  3. The study then looks at the bias that exists towards specific combinations of values used across parameter sets. For example a very profitable outcome in combination may be an Initial Stop ATR of 2 combined with a Donchian Period of 500 combined with a Trailing stop of 11xATR and a BE multiplier of 2xATR.  We therefore rank performance for each possible combination of value across Parameters from best to worst and then assess whether the top 50 of our total 3000 possible permutations also have values that lie within the preferred bias of each Parameter in isolation.  For example our Initial Stop ATR of 2 plots within the preferred bias range of Initial stop of between 2 to 4xATR, our Trailing ATR of 1 plots within the bias of the Trailing Stop in isolation of between 8 and 12xATR etc. etc. etc.

Results – Entire Universe – Individual Parameters

Table 1 – Entire Universe – Individual Parameters

Table 1 above summarises test results applied across the entire universe of 63 markets and consolidates performance results against each value assigned to the Parameters.

For example the entire test sample comprised 1.4 million trades which is a sufficiently large sample size to determine whether any favourable bias exists in the market data towards the key parameters used in this simple  trend following model.

Initial Stop Loss Values

The values of ATR used for our stop loss ranged between 2 and 10 ATR with step intervals of 2xATR.  The ‘Bias on GP’ column (Bias on Gross Profit before Trading Costs) looks at the median and mean results of overall performance outcome based on profit result and ranks the ATR settings from ‘most favourable for trend following’ to ‘least favourable’. There is a clear bias in the overall result demonstrating that ‘before trading costs’, the tighter initial stop loss is preferred to the wider initial stop loss.

So the good news here is that there appears to be a slight bias in our choice of ATR setting used for our Initial Stop loss….however the bad news is that this observed bias is insufficient to account for the total costs incurred for trading this entire universe the same  way using this simple method.

Note the impact of trading costs and in particular the SWAP costs (holding costs) when applied to wider initial stop losses that have (in general) higher holding periods as the initial stop is wider. The total holding costs alone are sufficient to defeat the overall bias in the series to produce negative expectancy with the Law of Large Numbers. 

Refer to the ‘Bias on NP’ column (Bias on Net Profit after Trading Costs) which re-ranks the series based on overall Net profit across the entire universe. You will see that the overall bias has been totally flipped to where the wider stop loss is now favoured to the tighter initial stop.

This flipping of the bias is a result arising from the impact of trading costs. The re-ranked bias now prefers those settings which result in fewer overall trades (those with a higher ATR initial stop) given the impact of trading costs alone.

This clearly demonstrates that the costs of trading must be factored in to our decisions when considering which markets to trade our trend following systems.

Trailing Stop Loss Values

This is interesting (refer to ‘Bias on GP’) at it suggests that a bias does exist in the choice of values used, but on a ‘before cost’ basis, there is a split in the bias towards shorter ATR trails and longer ATR trails at the spectral ends of the series. This just means that we need to unpack this bias more to define whether this bias relates to different selections of asset class or not. We will find that this riddle is explained by the end of this Blog post series. 

But now let’s look at the ‘Bias on NP’ (after cost basis). You will see that this subtle bias in GP is now dominated by the impact of trading costs to place the emphasis of bias towards the shorter ATR.  It is the dominant affect of increased SWAP costs of holding periods that starts to dwarf the impact of bias in the Gross Profit performance results.

Once again…you can see how the impact of trading costs is significant when dealing with fairly weak underlying biases that exist on a ‘before cost assessment’.

Donchian Period Values

Once again. there is a more jumbled bias across the universe in terms of ‘Gross Profit’ that prevents us from making any more definitive conclusions until we unpack this riddle more.

But note how on an ‘after cost basis’, the bias exerts its preference towards a longer look-back. This is easy to understand as the fewer trades with a longer look-back results in lower total trading costs…which is to be expected.

BE Multi

Not much can be gleaned in terms of bias in the series when looking across the entire universe. We need to delve further into this one in future posts.

Results – Entire Universe – Combined Parameters

Table 2 – Entire Universe – Combined Parameters Vs Individual Parameters

In terms of possible permutations using combined variables (of Initial Stop, Trailing Stop, Donchian Look-back and BE multiplier) there are a total 3000 different possible combinations. Given the small positive bias we may see on a before cost assessment which is dominated by the negative bias associated with trading costs, there are actually no profitable possible combinations that can be applied across the entire universe.

This therefore prevents us from adopting a ‘universal posture’ where we apply a particular  set of combinations to our Donchian Breakout System and cross our fingers that it is sufficient to ‘expect’ an outcome with overall positive expectancy at the portfolio level. The bias simply isn’t sufficient to allow us to adopt a universal principle across our entire universe. So what this means is that ‘unfortunately’ the bias in our retail world makes us have to adopt a more selective stance towards which asset classes and markets we decide to trade with simple trend following strategies.

Table 2 compares and contrasts the Top 50 permutations across Parameter sets against the ranked top 3 biased results from the Individual Parameter assessment conducted in the prior section of this post. Now while all these top 50 results applied across the universe do not possess positive expectancy after costs, we can at least evaluate where our model displays some overall beneficial bias (less negative expectancy than other permutations)

We can see that the Top 50 result in terms of Donchian Look-back period fall within the range of bias that exists in the Donchian Look-back when assessed in isolation. This is tending to suggest that a longer term look-back is desirous with a trend following approach when applied to this universe…but more on this later.

In terms of SL Multi….we can see that the Top 50 results lie slightly lower than the overall results of the bias of the initial SL when examined in isolation. As we progress in this analysis you will find that we can explain the reason for this slight discrepancy.

For the TSL Multiplier, the Top 50 plot within the preferred bias of the TSL in isolation as does the BE Multi.

Summary

Well this is a start and enough to consider in Part 1 of this series. Lots of information to take in and digest from this analysis alone.

As you can see from this brief introduction…we have our backs up against the wall in the retail environment when trying to find trading systems with a sufficient bias to overcome the costs of trading.

The Brokers have had a lot of fun at our expense in this world…but with a bit of due diligence in your assessment of this very uneven playing field, you can start to see that there are some biases that may be worth exploiting….but you first of all have to find them.   

In future Parts to this series we will start unpacking this universe into constituent asset classes to see whether some of these findings can be explained more thoroughly and conclusively by referring to the particular characteristics of each asset class in terms of their ability to provide a favourable context for trend following approaches.

As you can see with the start of this process….we are like the crafty gambler who arrives at the casino to first simply take some time to assess where the biases lie in the systems they are examining. Once we can determine this bias and confirm whether it is sufficiently material in nature to offset the costs of participating in the game….then we will be able to confidently define our trading universe from a quantitative and more objective manner than a more subjective assessment based on the different opinions we may hear from industry professionals.

Stay tuned for Part 2 of this series. As we start to reveal more of the findings…you will start to see method in this madness and you will also start to see how we start to turn the Brokers ‘house edge’ into ‘our house edge’….It is time we Retail traders start to re-calibrate that uneven playing field. 

Trade well and prosper

The ATS mob

 

Leave a Reply

%d bloggers like this: