Systematic Trend Trading

Hi R. For my purposes I use Dukascopy and Pepperstone to cross validate. No problems in data quality for H4 or D1 IMO.

For example here are two results taken from the same sub portfolio and time frame (1/1/2007 to current). One uses Dukascopy Tick Data and the other uses Pepperstone Control Points.

There is very little variation when dealing with longer term data on our particular strategies.

Stepping Outside the Noise to Hunt for Outliers. A Quick Proof

Attached Image

Hunting for Anomalies

Cheers

C

CTA Fund Performance Report - 30 September 2019 - Ouch!!!!

Attached Image

...more here :-)

Cheers

C

ATS Trailing Stop Visualiser

A Step by Step Guide Video Series to Portfolio Construction for Price Followers -Part 1 - System Design Principles

ATS PROGRAM PERFORMANCE– OCTOBER 2019 – APPARENTLY NOTHING!!

Attached Image

https://atstradingsolutions.com/ats-...ently-nothing/

Fred steps through Part 2 of our Video Series on how to Construct powerful diversified systematic portfolios. This step discusses the automation process of system development.

Inserted Video

Part 3 of a video series that describes the process that we undertake to construct powerful robust systematic portfolios for the Price Follower in the FX/CFD retail trading universe.

Inserted Video

Hi Copernicus,

Thanks for your detailed answer. I thought I would get a notification if someone mentioned me in a reply, hence I didn't look back to the thread. I think I got confused with the way discord works. Anyways:

- So you look at a (much) bigger picture than me in terms of time frame. So a first thing that I find interesting / confusing, is that given the TF of H4, a 4 year period corresponds roughly to 24 years of D1 in terms of data points. I would honestly assume that it's more than enough to get a valid expectation of the approach. The equity curve that you post shows otherwise, but still perplexing. One thing you said that struck a chord with me, is that you're essentially fishing for outliers, the rest being boiled down to trading randomness. I've tried many different trend following approaches, usually on H4 and D1, but with an in-sample period of about 4-5 years. And every time it didn't give a positive slope over a few pairs, I'd ditch it and move to something else. Maybe part of the solution is to expand the testing period.

- Which brings me to a second point: say you're trading solely this type of system, over a relatively large portfolio (28 pairs & some commodities). Given the equity curve you mentioned, you would have to be comfortable with 4-5 year periods were the account would be stagnating. That by itself seems quite difficult to overcome, or am I missing something?

- Regarding the individual results that you posted, quite a few pairs had no activity. I assume that meant no signals came up on that pair, but it seems surprising given the time period.

- What software / platform are you using for the portfolio?

Hi B....I tried the like on your post...but now I am commercial it doesn't work anymore...so here is another

I will respond later this weekend to your gem of a post....and hopefully might be able to blow you away with a bit of empirical reasoning from the data. You are right when you talk about long periods of stagnating.....and that is exactly what we do find with this technique........and that is why you need to zoom out and look to the long term say 10-20 yrs to see the wealth building returns.

This technique is all about applying a small 'enduring' edge that with compounding through extensive diversification to deliver sustainable wealth building returns.

It is about mimicking the techniques of the best FM's in the world with a validated track record as opposed to a predictive punt with leverage where you close your eyes and hope for the best.....which we know is a common story in the retail space.

There are no validated long term track records of the latter 'retail' story.....but there are a huge array of validated long term track records from the diversified systematic trend following camp of Professional FM's.

In fact this Index of 37 FM's in the trend following space have all outperformed Buffet over this time period in terms of risk adjusted return (MAR ratio). It was reported first in 2010....but it continues to today across the entire Index of 37 x long term Trend Followers referred to as the TF Global Index (37).







Cheers C

Hi B....back to your queries. It's been a long weekend and an exhausting one :-)

Yep D. We let the data decide how far we look back when trying to identify that zone where outliers dominate overall price movement. At the moment no matter what timeframe we use.....this zone appears to be in the range of about a 250 to 500 D1 bar lookback. It seems a really long lookback but the data is pointing to that. Markets are pretty efficient these days or mean reverting in nature. You need to get outside the normal everyday market churn to achieve great R:R.

The equity curve at the portfolio level is a result of pretty extreme diversification. You will see the step ups in periods of uncertainty such as 2008/2009, 2010/2011, 2014/2015 and perhaps just now...which have been where outlier impacts have dominated and despite the intervening drawdowns have been more than sufficient to compensate for the losses of less turbulent times.

That's it mate. You will find that if you try to catch trends during more noisy (less exotic) conditions, only a very few instruments will work.

Effectively in our opinion there are 3 types of trends. Random trends that have no enduring substance (which are found in the normal distribution zone of returns) , trends that are part of larger mean reverting cycles, where markets may oscillate about temporary equilibrium's (which are found in the peak of the distribution and relate to specific market conditions) and anomalies that are a result of markets transitioning between equilibrium. This latter form of trend is what we are after and can be found in the fat tails of the market distribution of returns. It is the outlier that actually is common to all liquid instruments.



We can demonstrate this with a test we did using a simple Donchian breakout system over 1/4 million trades across 15 randomly chosen FX markets using different conditions that progressively increased the look-back period.





So what this test concludes is that the 250-600 day look-back offered the most profitable results of the series with low Pwin% but very high relative R:R.....and as you shortened the look-back they clearly became less profitable almost linearly as you reduced the lookback. Interestingly as you reduced the lookback the Pwin% increased but what was evident was that the R:R significantly decreased hence the big drop in profitability across all markets.

There were still a few winning solutions in the shorter lookback but there were no profitable results across the entire 15 markets with the short lookback.

This test supported the assumptions of the model and strongly suggest that in more extreme zones, price is more extended in nature than in less extreme zones. This makes sense when you look at any very large long term trend. You see that the daily bars get more extended as the trend matures and gets into a capitulation zone where outlier moves are very evident.

Because the trends in this extreme zone offer this degree of price extension....the R:R can be very large indeed. It makes us therefore far more selective in the class of trend we want to catch but we diversify very widely to catch as many as we can. They are far more frequent than many realise.


Yep mate.....I am comfortable with being patient. The rewards are so great to the PL if you are. We simply cannot afford to miss the outliers and we never can predict them so we need our systems to be turned on 24/7 protecting our capital at all times...and we simply lie in wait with broad diversification across markets. We treat this as a long term wealth building exercise where we are in for the long haul....as opposed to a cashflow generation exercise. Those after a reliable cashflow succeed over the short term but are likely to lose in the long term.

It is the markets that determine your fate in this game and you can never predict with certainty what they will do. When they decide to get nasty....it is the killing fields for those seeking cashflow.

Just MT4....some Excel spreadsheets and an Access database. We do use Quant Analyser occasionally to quickly import MT4 trade results and produce some report outputs but we are slowly reducing our dependency on any off-the shelf-solution.

Great post Copernicus! FX markets are tending to become less volatile lately (drop off in Daily ATR), do you see that as a problem going forward?

Hi R

It just means we spin our wheels in stagnation while vol is low. We have had protracted periods of low vol since 2009...but then occasionally something snaps like in 2010 and then 2014 and more recently in July 2019 which catches the central banks off guard. Because we have such distally placed entries waiting for price to do something 'significant' not much happens at the moment. We occasionally get a false breakout or two (snags) where we get a little uppercut now and then....but when things heat up usually when the S&P500 takes a noze dive......it's gonna be a turkey shoot :-)

The snaps have become progressively more significant post GFC....so looking forward to another one.

Tip of the Day

Before you try and win at the casino you need to decide if you can beat the house edge. The first step you need to take before committing to any trade is first understand your table. If there is no bias that gives you that necessary 'edge'....then quite simply you are punting. It is no different to medieval quackery.

Screen time does not equate to a method of honing an edge. Unless you can demonstrate that an edge exists in your systems long term track record through some form of empirical testing ....then it's all just smoke and mirrors.

In a 'fair' coin toss game with 10,000,000 participants where two participants each bet a $1 stake and the loser leaves the game and gives his $1 to the victor.....then at the end of the game when only 1 participant is left, they win $10,000,000. That participant has not deployed any skill....it is luck that simply determines that outcome.

However if you play a game using a biased coined....over a very long data set, you can quantify that bias within a confidence interval and deploy that very fine edge to turn it into something definitive and meaningful (within the bounds of certainty that random walk across that biased series allows for). There is no guarantee....but there is a bias in the coin flip series where if you bet very small (with limited leverage)....that bias can be turned into a large wealth pot over the course of time simply through the compounding impact of that small edge with limited leverage.

To turn voodoo magic into a process with rigour you first need to empirically test the validity of your model. You must deploy the scientific method in this game if you want to be a long term survivor.....otherwise you will be rudderless and permanently lost and subject to the whims of what randomness can deliver.

Always remember that markets are very efficient....particularly the Forex market. The vast proportion of your track record can be attributed to luck alone...however the important bit of that track record....which is very small indeed....that allows you to have a greater chance of surviving this 'Machiavellian' market is the very fine bias that may exist in that series. No visual technique will be able to identify that bias when trading on the right hand edge of the chart. It is the results of your empirical method from the past that is your most reliable guide for that uncertain future that lays ahead.

Just as a casino deploys quantitative statisticians to validate their gaming models to turn speculation into a business enterprise....so too must the trader seeking to capitalise on any house edge that the market may deliver from time to time to become a proficient speculator as opposed to a gambler. Due diligence using the empirical method is a must for those traders who seek the confidence that their models actually represent the 'reality' and are more than just smoke and mirrors or a curve fit outcome of no substance.

[quote=Copernicus;12609742]Hi B....back to your queries. It's been a long weekend and an exhausting one :-) {quote} Yep D. We let the data decide how far we look back when trying to identify that zone where outliers dominate overall price movement. At the moment no matter what timeframe we use.....this zone appears to be in the range of about a 250 to 500 D1 bar lookback. It seems a really long lookback but the data is pointing to that. Markets are pretty efficient these days or mean reverting in nature. You need to get outside the normal everyday market churn to achieve great R:R. {quote}

Hey C,

If the "outlier zone" for large asymmetric price moves is 250-500 D1, it would seem that a mean reversion system could be developed to take advantage of this long stretch of time between the large R:R trades...

Dave

Great point Dave :-)

We are going as far as we can with trend following to see what we can get....but it would be stupid if we didn't also look at mean reversion below these "outlier thresholds".....but if we use it....We just need to ensure it doesn't compromise the portfolio result. We will be looking at this at the tail end of this exercise.

Best of all worlds.

Cheers D

To give an idea where the market adopts more unidirectional price moves. you can see this bias in big data sets.

For example using a universe of 60 markets the bias lies at around a look-back of 500 to 800 daily bars plus. This is found with a data sample of 15 million trades using a very simple trend following system. What this means is that you need to trade the higher timeframes to capitalise on the edge that exists with the market data.....otherwise your result may as well be a random one.



This range is almost a common range across asset classes. The trade sample sizes here are sufficient to make pretty strong conclusions.

Forex



Equity Indices (via CFD's)



Futures (via CFD's)




The best trending markets are obviously Futures. Advantageous price extensions for trend following clearly dominate in the highlighted zones of bias and for futures, the look-back is much shorter than Forex or Equity Indices. In other words Futures markets are far less noisy and mean reverting in nature than Forex and Equities. In fact in Futures some of the rules such as the look-back are universal in nature (provide positive expectancy) across the majority of instruments. Not surprising really...but this analysis just validates that assumption.

No wonder most Professional Systematic Trend Following FM's occupy the futures segment. The data clearly displays why it is beneficial for TF firms to use this asset class.

So if you want to apply trend following to FX or to Equity Indices....you must adopt a 'selective' mechanism such as a rotational strategy to only select the best trending markets (which appears to change over time). We are also investigating whether trading positive SWAP only is sufficient to turn this asset class into one with a universal application for Trend Following.....but as it currently stands, we need to be selective in the markets that we trade in Forex.

In commodities at least over the past 20 years, the universal rules means that just by applying trend following principles, the market bias is on your side.

The feeling of having a definitive bias on your side does wonders for an 'uncertain' mindset. Clearly as a trend follower you would be foolish not to include CFD's and ETF's into your universe across futures markets and commodities.

Here is the Universe we tested on and the available data ranges.

Traders Outpost Podcast Ep01

Hi C,

Going back to your previous post where you show the outcomes of your trend following strategy. If I take for example your first picture, which has USDCAD included. There is no scale so I can't really compare, but with the dates I chose a particular period in order to compare with what I have. So I take the period that really pumped up the equity curve, which seems to be 2014-2017.

So when I choose that time period for the channel breakout strategy, and use your 9 ATR trailing stop, I get -5%. If I use a 3ATR TL, I get +16. Better, but doesn't seem as explosive as you chart shows. Again, I don't have the scale so not really sure. But one thing that's sure, if I take the longer time period for that same pair (2008), my expectancy is negative. So very different from yours.

With all those backtest results that you show, is it with the channel breakout strategy that was talked about? Did you apply the same settings for all pairs? If we're talking about the same strat, I don't understand how our results can be so different.

Thanks for you in-depth posts.

Morning B. A late start for me today :-)

The USDCAD example was taken from a strategy series called Uranus.....which is a Bollinger Band breakout system combined with a trend direction filter used on H4. So it should not be compared against the simple Donchian breakout on D1 which we have undertake extensive testing on with this post.

The USDCAD result was derived from the following settings where we used R and not ATR as a method of trail in this example.



Here are the equity curves from Dukascopy and Pepperstone. They are virtually identical if you mapped these to commence at the same starting point.

I hope this helps a bit mate.

Cheers

C