Hunting for Anomalies
We have spoken at length in our prior posts about how the Price Following mindset owes their fortunes to the outlier. The most successful investors and traders on this planet owe their ultimate success to the outlier….which you can’t predict….but how therefore can you catch them? This post digs deeper into this little chestnut.
If you take the time to understand your trade history as a Price Follower, then you quickly realise how it is the unpredictable right tailed event that lays the foundation for your long term track record. As for the rest of your trade history……well that vast array of effectively random outcomes are the necessary results of the grind that occurs while you patiently wait for the unpredictable right tailed event.
Now if you have been obeying your mantra of ‘cutting losses short’ at all times, then that grind effectively hovers around a neutral result but with a slight inevitable deterioration during periods of doldrum when market transitions are not prevalent.
So I frequently get queries posed to me from traders seeking to emulate the Price Followers…… that it is very difficult to find simple trend following strategies that have multi-market potential …….. and also questioning the assumption that all liquid markets display these market transitions. They would like to therefore understand why their trend following strategies only work on a few markets…as opposed to many of them.
The answer to this little nut of a problem actually relates to whether or not your strategy is trading within the ‘normal zone’ of day to day market condition or whether it focuses it’s trading activity to the more ‘exotic zones’ of non-normality.
In understanding the reason for this statement it is useful to think of the mix of participants that reside in different market conditions and the influence that their activities have on overall market behaviour.
Within normal trading conditions, there is a plethora of different types of participant doing their own thing. The array of different methods used to enter and exit the market leads to significant noise in the overall market condition. Sometimes there is an overall dominance in the mix of a particular form of mean reverting behaviour which leads to predictive oscillations around some historic equilibrium point. Sometimes there is an overall dominance of trend following behaviour where uni-directional price movement starts to dominate within this ‘normal zone’ of market activity.
Now despite the fact that we can clearly see trends even within normal trading conditions, the reason we avoid trading while residing within this highly competitive zone is that the impact of this trending behaviour is far less dominant than what exists in more ‘extreme’ zones. In these extreme zones outliers have a major say in the total market condition. So if you want to snare trends during these more normal and efficient market conditions, be prepared for noise and mean reversion to lead you to many whipsaws and a steadily deteriorating equity curve.
It is only during more exotic ‘extreme’ market conditions that participant behaviour becomes more coordinated in it’s impact on overall price movement. What previously was noise arising from the interference effects on overall price resulting from a vast array of different trading behaviours now becomes much more coordinated where a single form of trading behaviour begins to dominate the overall price movement. Efficient markets start to lose their grip in these zones and become quite inefficient. The anomalous unidirectional price move dominates the noise and mean reversion tendency in the outlier zones. So if you trade trends in this zone, your whipsaws are generally less (according to the Law of Large numbers rule ……aka many trades rule)
For example during market capitulation phases, uni-directional price moves can get extreme, and this is due to the fact the bulk of predictive participants are doing the same thing……and that is exiting their rapidly deteriorating positions at all cost.
So enough of the diatribe….let’s have a look at what the data says and see whether this assumption holds any weight….and look at it’s implications.
For this example I will use a very simple Donchian breakout system using the preferred rules of a Price Follower:
- Trades in the direction of overall price bias;
- Applies filters to avoid trading during the everyday market churn and strike during more exotic conditions;
- Applies a momentum filter on entry to ensure that when the Donchian breakout is triggered there is short term momentum backing the price move.
- Cuts losses short at all times by applying a trailing stop condition from entry through to exit; and
- Lets profits run with no profit target applied where our trailing exit simply ensures we exit the position at some time in the future when triggered.
In this example we have codified this simple system to allow us to quickly assess different parameter settings based on it’s application across a universe of 15 liquid Forex instruments over a 20 year backtest horizon on the Daily timeframe:
Now in defining the normal zone versus the outlier zone, we will use the look-back period of the Donchian Channel itself to give us a feel for whether the entry signals lie within more normal conditions versus more exotic conditions.
For example a 50 period look-back places a channel at the high or low of the last 50 daily bars. This fairly narrow channel would be regarded as representing more ‘normal’ market conditions where price is oscillating within this channel. On the other hand, a 400 period look-back provides a very wide channel in which our assumption is that for price to reach these more extreme zones, something more than a random result is likely to be backing it.
So lets see what the results conclude based on a 20 year optimisation test across 15 different liquid instruments. In this experiment we have 1,444 different possible permutations of result in our test. These 1,444 possible permutations are applied across 15 different liquid markets. We then need to collate combinations of possible result to see what is going on in the entire large data set of 15 x ,1444 = .21,660 options. This set is large enough to see if a bias in outcome is present.
We sort by the MAR ratio (CAGR%/Max Draw%) as a basis to define the better results in terms of risk:return relationship for each return stream and see whether there is a clustering of settings.
Here are the top 45 results of the analysis based on MAR sort.
- The average look-back of these 45 different permutations was 402 days;
- The average initial stop applied to all these 45 options and worst case adverse loss was 2.1 ATR to all these permutations;
- The average trailing stop was 12.0 ATR allowing for significant open ended profit potential.
Conclusion: 44 of the 45 options produced an overall profitable result across the entire 15 markets over the 20 year test period. The average profit over the 45 options was $1721. The best scenario was profitable across 7 of the 15 markets with a worst case loss of ($236) for a single instrument and a best case profit of $1290 for a single instrument (demonstrating strong positive skew).
Here are the bottom 45 results based on MAR sort.
- The average look-back of these 45 different permutations was 59 days;
- The average initial stop applied to all these 45 options and worst case adverse loss was 4.0 ATR to all these permutations;
- The average trailing stop was 8.0 ATR.
Conclusion: 45 of the 45 options produced an overall losing result across the entire 15 markets over the 20 year test period. The average loss over the 45 options was ($3,057). The best scenario was profitable across 5 of the 15 markets with a worst case loss of ($685) for a single instrument and a best case profit of $662 for a single instrument.
Summary: The data speaks for itself. If you sort by MAR across the entire possible set of options you can see that best results have a higher look-back with worse results having a shorter look-back. In fact with the progressively bigger data set, there is no single best set of combinations, but rather a general assumption that rises to the fore. That is the general assumption that the longer look-back, the better the overall result but the fewer trades in your sample. To overcome this you need to broadly diversify across a broad set of permutations with a broad look-back to lift your trade frequency as opposed to invest all your effort in any particular combination that might be successful. We are the casino in this example tilting the odds of success in our favour by diversifying widely to capture the overall bias grounded in this assumption.
Overall profitability therefore in a multi-market system context requires you to adopt a longer term look-back with your applied filter that attempts to separate the normal zone of day to day price action from the anomalous zone. This is symptomatic of the fact that for successful harvesting of the bias that exists in the data you need to step outside the noise of the everyday market churn. You will not be able to reap the long term benefits of the outlier by applying generic simple trend following strategies across multi-markets unless you are prepared to step outside the normal zone into that area where outliers are known to dominate the overall price action.
So if you have designed a breakout system in this universe with the expectation that a 50 period Donchian was sufficient to apply in a multi-market context…..then you would be wrong. You would be required to cherry pick particular markets….which is always a hind-site measure when looking for favourable anomalous impacts. For example in the 50 Donchian look-back examples you still find profitable markets over the data series…but you need to therefore be selective and just hope that those particular set of market conditions for that particular profitable market continues on into the future.
Now the sad tale of this is that your selective choice actually makes you a predictor as opposed to a price follower. You simply cannot afford to cherry pick as you revert to a predictive mindset in doing so.
Now most retail traders would not consider such long look-backs as 400 days or so as the demand for instant gratification makes them look short term in their strategy design. This therefore biases them towards a predictive mindset as such short look-backs make it virtually impossible to harvest the fruits of the outlier given the overall noise in their short term view. Learn from the best FM’s in the world when they say that cherry picking markets ends in tears in the long term. There is a reason why most systematic diversified Trend Following FM’s have progressively stepped out towards the longer term to harvest alpha. Markets are far more efficient than they were in the past. The short look-backs that worked in the past no longer work today because of this feature.
Retail traders tend to work off gut judgement or hearsay. All we can say is that there is a solution to trading these markets under systematic diversification….even in FX alone…..but to find your edge you cannot rely on hearsay and must do some due diligence and see what the data says about your assumptions.
But as we can see in this brief example we have a profitable solution provided we apply a bit of rigorous analysis to the data….. but to reap this profit you need to have many diverse systems all with long look-backs over a long data set to benefit from this apparent edge. This therefore necessitates diversification and a method to compile these solutions as a comprehensive portfolio.
Trade all your liquid markets the same way but ensure your entry signals are located in the outlier zones. Remember an anomaly is related to a market transition as opposed to a predictive pattern. They can occur any-where and any-whence in all liquid markets.
There is a finite and ephemeral edge in predictive techniques but an unbounded (potentially infinite) and enduring edge in Price Following techniques. It is the unbounded profit potential of Price Following that makes this game all worth it in the long term. Protect your capital at all times but leave yourself open ended to what the market can possibly deliver. Let the market dictate your results as opposed to your clever ‘predictive’ system design.
Trade well and prosper
The ATS mob