Diversified Trend Trading Approach

Why is it everytime I post a moderately singeing comment I get bombarded by these textbook responses. Oh Forex Factory, don't ever change.
If we are talking about funds like Dunn, Abraham and Chesapeake I don't actually think there are any 'special secrets'. Its most likely vanilla trend-following with some flavourful sauces.

I was impressed with Andreas Clenow's ability in his book to mimic equity curves (in specific detail) of the Big Boys by playing around with various investment universes and systems. It made me realise that with a bit of crunching it is not rocket science, but what really grabbed me was the ability for FM's to carve out unique equity curve solutions that would be attractive for particular classes of investors looking for a particular performance complement to their existing portfolio's. What is evident in the FM industry and in particular with the big funds in the trend following space in more recent years is that they are progressively getting more and more diverse returns associated with their need to niche in various ways by offering distinctive equity curves with varying correlations to the wider market to attract investment funds.

Did his research yield anything insightful? Im under the assumption that performance is linked to the trends in the traded instruments.

The point of his exercise was to demonstrate that many of the funds with 'Wizard" status can be replicated using a simple trading strategy or combinations thereof and the essential ingredient to their success was attributed to diversification and volatility position sizing as opposed to actual trading method .

By being able to mimic their equity curves with acute accuracy, he found that performance differences were attributed to the nature of investment universe traded (where he used 5 different specific sets of investment universes with varying exposure to different asset classes) and it was necessary to adjust the position sizing formula used which was necessary to match the volatility level of the fund. Interestingly he used a single timeframe in the replication to mimic results but a variety of different trend following and counter-trend systems in order to capture short term trends, medium term trends and long term trends.

Here are some of the results which are extracts from his great book and demonstrate quite clearly that he knew what he was doing in his reverse engineering.

Interesting, this corroborates my current understanding.

Am I correct to say that the main differences in performance between the funds have more to do with differences in their portfolio composition rather than anything else?

And what are the '5 specific sets of investment universes'?

Yep I think you are right. Portfolio composition as far as my take on his research accounts for the lions share of performance attributes. Here is a snapshot of the different universes he was meddling with. The graphs above for the FM's state which universe he uses for them.


Equal Weighted Investment Universe


Large Universe


Large Universe, reduced equities (as highly correlated)


Commodity Heavy Universe


Financially Heavy Universe




Comparison

Attached Image (click to enlarge)

Good stuff, thanks for posting.

Some great recent podcasts on Top Traders Unplugged with Anreas Clenow. Definately worthwhile listening to.

1. Start with this.

2. Then this.


Enjoy

C

A New Addition to the Thread - Fully Automated Systems and System Diversification

As mentioned previously I have been working on a project in MS Excel that is tailored for my particular method of back-testing across a diversified portfolio and a diversified range of systems.

The intent of the project is to develop a range of systems that can handle an extended array of market conditions (at least those conditions experienced across global markets over say the last 15 years) and hopefully (crosses fingers) give a good challenge to the successful CTA's out there who have outperformed alternative investments over a similar period.

I will be using the TF Index (a hypothetical index of the top trend following FM's) described in Post 550 as a comparative measure to assess the performance of each system and the ultimate composite suite of systems that will emerge from this process. Those composite system's that exceed the performance of the overall TF Index will be then chosen as candidates for forward testing and inclusion in the monthly performance results that I post on this thread.

Once a system suite is selected, to be fair, I will then be adjusting the returns to include standard fund management fees such as IM fees, Performance fees, Custodian, Administration and audit fees) so we can directly compare the performance of these suite of systems against the FM's on a level playing field.

Now this is a work in progress, so you will need to be patient with this process. The procedure will test a vast range of systems in isolation across a diversified portfolio for the primary purpose of producing equity curves that will be used to ultimately construct a very robust trading system that avoids curve fitting.

As new systems are trialed, the results will be posted on this thread and hopefully will prove useful to those interested in following a similar path into systematic trading. Along the way, some useful info will emerge that might challenge conventional viewpoints of members on this forum....but hopefully this will prompt others to perform similar tests to validate whether there is any merit in these findings.

This thread will therefore house a large number of performance charts for varying systems, most of which will go nowhere. So don't get frustrated. The intent is to only use those equity curve contributions that are useful to flattening the variability of returns of a composite suite of systems. I hope this all makes sense............

Now onto the good stuff.



The Investment Universe

The following selection of instruments will comprise my total investment universe. Unfortunately it is limited to my brokers offering but it sufficiently spans across asset classes. We can adjust this investment universe down the track if we want to focus our strategy towards different asset classes such as equal weighted portfolio or commodity focused portfolio etc. At the moment it is fair to say that this universe is biased towards forex, so at least most members on this forum should be happy.



Below is the performance results of the TF Index from 1 Jan 2000 to 31 Dec 2015 which will be used for comparative purposes.



The next post will make a start with our first system - A Classic Donchian Breakout System.

Regards

C

The Simple Donchian Breakout System

Note: All the systems described by this project will be using the D1 timeframe......simply for the purposes of this exercise.

The project will only focus on classic simple systems as overly complex systems are difficult to analytically interrogate for this exercise.

The project starts with one of my favourite systems, the Donchian Breakout which sits in the suite of classical mid-term range trend following systems. The System rules are as follows:
1. Enter on open if prior bars high/low was higher/lower than 100 period Donchian;
2. Set initial stop on entry at 2 x ATR (14 period) with fixed trade risk of 0.25% of trade capital;
3. Exit on open if prior bars low/high was lower/higher than 20 period Donchian.

The non-scaled results are as follows:



Hopefully you can magnify the attached chart to clearly see the results.

Salient points are as follows:

1. Look at the very low Pwin of 24.3%. This is the reality of mid-long term trend following performance since 2000 which has been a particularly tough period for trend followers in general apart from a few stellar performance years.
2. The # Long signals and # Short Signals indicate that only a handful of these signals were taken by the strategy. Once a trade is committed, no further trades are taken on a single instrument to avoid overly aggressive correlation and portfolio risk exposure.
3. Look at the total costs of transactions. Over $100K of profit (almost 40% of gross profit) was allocated to commissions and SWAP/Interest Holdings Costs. This is a fact of trading that is cleverly concealed by MT4 results.
4. Total (non-reinvested) return over the period was 103.45% contributing a miserable 4.54% PA of compound annual growth (CAGR). Later this will be scaled and profits re-invested so don't get to disheartened by this result.
5. The maximum drawdown for this effort was 11.68% for the period. I am finding as a general rule for long range testing that you can probably as a rule of thumb assume twice the drawdown of the average annual return produced.
6. Profit factor for the period was 1.42.

Despite the meager overall results we need to put this into perspective. There is considerable variability of returns with trend following strategies of this type. It is just the name of the game. Have a look at the annual performance results below.

Across the entire portfolio of 40 instruments, 25 were profitable with 15 being unprofitable. The right hand side of the chart reveals profit performance for each instrument. While it would be tempting to turn off those poorly performing instruments using the assumption that they just don't trend......what does not trend today almost certainly trends tomorrow. We are not in the game of predicting here so we just need to suck it in and roll with the punches.

So how does this single system perform against the TF index when scaled to achieve a 20% drawdown and where profits are reinvested ?



The chart above compares the D13 system (orange) against the TF Index (Yellow) and includes key performance metrics such as the strategies correlation to the S&P500 TR Index, and TF Index. Clearly the low correlations suggest that the TF FM's are using a range of different systems to that described in this example.

It has been a tough time for trend followers since 2000, but still such simple strategies as these give the S&P500 TR Index a run for their money. Clearly we need work to do to bolster performance to get closer to the TF Index, but even as a singular system, results can't be sneered at with CAGR of 6.65%, a max draw of 20% and last 12 month return of 52%.

It is important to note that all positions were closed at 31 Dec 2015 to gauge total performance over the period which results in the very large return of 41% for December. What this suggests is that a lot of the FM's in the trend following space are carrying significant un-realised profit in their portfolios as at December. Time will tell if this conclusion is correct.

On Thursday evening I will be adding additional systems to the suite to see if we can lift performance of our portfolio and get closer to the big boys.

Cheers

C

Interesting how your equity curve for D13 carries the same characteristics as some of my results.. returns plateau around 2010 though I manage to get positive returns in all years since 2002.. but.. I haven't incorporated some of the costs of trading and I'm not as diversified as you yet which may change my results once all asset classes have been looked at..also interesting is your worst performing instrument Is EURCAD.. its mine too, and I've tested this beast with completely different models... its one of the worst performers with my discretionary trading too..


Cheers C..

Composite System Objective
As our mission is to identify composite systems that serve to flatten the equity curve and reduce the variability of returns with the core strategy/ies of the system (namely trend following systems), a useful way to achieve this is to use Van Tharpe's matrix of market conditions to classify which market condition each system best responds to. By doing this you ensure that your ultimate suite of systems can respond to the complete set of market conditions to make it as robust as possible going forward.

Generally speaking, using trend following systems as a core strategy is a wise move given the unpredictable nature of markets in general and their propensity to display extreme volatility at unpredictable times when the elastic bands snap from central bank regulation. Alpha crisis events might be fewer and farther between, but their intensity is likely to increase if market manipulation continues to be a feature of the global economy. When the elastic band breaks you want to be on board the trends as they are likely to be fast and furious events. To lie in wait for extended periods, you need to invest in alternative strategies to take the heat away from natural drawdowns associated with trend following systems. You may note that trend following in the FM space is becoming more diverse in the range of systems employed. This is attributed to the fact that since 2010, trend conditions have been few and far between and may be attributed to central bank intervention practices.....anyway, this is digressing a bit.




Above is a matrix of possible conditions and the coloured zones represent those zones in which we are currently responding to with the systems employed. Obviously the medium term Donchian system described in the prior post is suited to trending markets so should plot somewhere within the up/down and quiet/volatile range of conditions. I have classed this system as belonging to to the quiet market condition as the exit condition of a 20 period Donchian suits less volatile markets. If I was seeking a trend following system to fit within the volatile range of market conditions, then I would be focussing my efforts to those strategies that enter into trend with a broad trailing stop or exit mechanisms that gives the strategy room to breathe.

Now that I think I have a suitable strategy for two market conditions (yellow) being the bull or bear low volatility conditions I am now looking for a system that performs in sideways markets. The following system that will be described is a classic mean reversion system which is configured to respond to sideways volatile markets. To respond to these conditions, we are after a mean reversion system that operates at extreme ranges of sideways volatility away from the mean of price movement and we want to give room to breathe for price movement (given the volatile conditions) and a profit target to reflect the fact that unlike trend following, mean reversion systems have a finite extended nature within a channel range.

By selecting systems that respond to varying market conditions, you are in effect ensuring that each system is relatively un-correlated with each other. In composing a suite of systems to flatten the equity curve, the name of the game is selecting un-correlated systems to achieve this. What you don't want is additional systems that exacerbate drawdowns even further when you merge them which prevents you from scaling up to improve composite system performance.

The Bollinger Mean Reversion System
The rules for the following simple mean reversion are as follows.
1. Enter on open if open is less than/greater than prior days open and is less than/greater than the upper/lower Bollinger Band of 2 std deviations (2Sigma) of a 10 period SMA. The 2 sigma BB is designed to operate within volatile markets giving as much room for volatility as possible.
2. Set stop at on entry at 3 x ATR (14 period) = 1R with fixed trade risk of 0.25% of trade capital. Note that more room is given to breathe on this method as opposed to the D13 system hence the wider ATR.
3. Take Profit Target of 4 x R.

The non-scaled results are as follows:



Now before everyone jumps up and down about the performance of this system it is important to note the intent of what we are trying to achieve here in trading a diversified portfolio of systems. We are after equity curve profiles that cumulatively add to performance of the core strategy (trend following) without significantly increasing risk exposure. If you closely compare the equity curve above with that produced by the D13 system, you will be pleasantly surprised and note either by correlation or by visual matching that when one system is in drawdown, the other system is performing. As a result, when you merge both systems and re-weight accordingly you can achieve improved performance with no or limited addition to overall portfolio risk exposure.

I would also suggest as a dose of reality that mean reversion systems are far more difficult than say trend following systems to stand the test of time and produce positive contributions to equity over the long term under a range of market conditions. What you will find is that during times of stress for trend following, mean reversion systems assist greatly in reducing overall drawdowns but on their own, are not worthy of representing a core system in any diversified portfolio (in general).

Looking at the stats you will see a Pwin over the period of 24.3% which is a realistic assessment over an extended timeframe of systems of this nature. Of course during periods of mean reversion, these ratio's shoot upwards of 50% and typically range between 50 to 60% win rate (which is higher overall than trend following systems).

The system un-scaled contributes a total $62K (or 31.12%) to overall portfolio performance with a measly CAGR contribution of 1.71% over the period (probably in the same range that a term deposit over the same period could achieve). The total drawdown for the period was 13.46% and of the 40 instruments, 27 were profitable while 13 were not. However despite these stats, this system could be a worthy contributor to the system composite.

Now let's have a look at the scaled result and the performance with the reinvestment of profits.



The chart above reflects the performance of the scaled B1 system. You can see than this mean reversion system under-performed against the S&P500 TR Index but was fairly flat in it's overall performance. The low variability of returns is going to be a big bonus for the system composite.

For the next post we will merge both systems and re-scale accordingly to see the combined result. Hopefully the process will see overall performance of the composite system improve beyond that delivered by D13 alone and track closer towards the performance of the Big Boys.

Cheers

C

Blending D13 and B1

Below is the blended un-scaled result of simply merging both systems together using the same portfolio.



Hallelujah brother. Performance is improving with limited variation in overall drawdown to the D13 system. The central reason for this result is the blending of two disparate systems of low relative correlation. When one system is in drawdown, the other system is performing. The blended result is therefore far more robust in handling a broader range of market conditions than a single system on its own.



Annual performance results also show less variability of returns while still capturing some great returns in 2006, 2007, 2009, 2013 and 2015. Poor performance results in the other years is also diluted a a blended result.

So let's see what we can do when we scale the composite system to achieve a 20% max drawdown and also reinvest profits back into the system.



Have a close look at the result and compare the equity curve of the blended system and the stats against D13 and B1 in isolation. We have managed to significantly bolster overall performance (lifting CAGR to 8.5% PA) without compromising the variability of returns or leading into deep drawdown territory. The result is seeing the equity curve start to lift and flatten providing a progressive improvement towards the performance returns of the CTA's. On it's own this blended system will outperform a number of CTA's already tracked in the monthly performance but we still have a way to go in our blending as we need to fill the gaps in Van Tharpe's matrix of market conditions to improve results further and head towards the optimal equity curve.

The correlation between the blended system and TF Index is still un-correlated meaning that we are yet to find the major contributing system that mimics the returns of the TF Index. Obviously our search now needs to head into Trend following systems that operate in volatile markets so a good place to star is to investigate those moving average cross systems designed to target the more volatile times. Following that we will need to find a good sideways low volatility system. Note that we will also need to look at some counter-trend variations and see how they contribute to or detract from overall performance.

Referring to the blended equity curve, areas of weakness still exist and relate to market conditions during 2000 to 2006 before things started taking off. If we can incorporate systems that performed well during this period we will be able to reduce the drawdown and significantly accelerate portfolio performance which is the aim of this game.

As we strive for additional performance in the equity curve, the harder it gets to achieve, unless we can find a system that specifically targets the max drawdown without exacerbating the overall variation of returns..... but hopefully this process provides an insight into methods used to improve portfolio returns across a variety of market conditions. Don't fall under the impression that this process is easy to achieve....it's not....but it may give you ideas on how to accelerate performance without having to enter high risk territory.

Now that I have explained in detail the processes of system blending, it will not be necessary to re-iterate this in future posts so I will be simply posting performance charts and allow you all to draw your own conclusions.

Regards

C

PS. Note that the B1 system is turned on all the time. Some may suggest that when one system is turned on, the other system should be turned off (eg. by using an ADX filter to trigger when we are entering trending conditions versus sideways congestion periods). However this solution does not address the central feature of what we are trying to achieve.....and that is to reduce the drawdown of the core strategy. If the B1 system was turned off when the D13 system was turned on, this will not assist drawdown reduction. It may result in improved overall performance but will not attack the variability of returns in the manner we are seeking. Furthermore, given we cannot predict the future, all systems are turned on all the time as you never know when market conditions change.

Yep Ja. The EURCAD has been a dog of a performer for trend following since 2000. You should take a look at post 593 however as this reveals that EURCAD did quite well with mean reversion over the same period. It's has been a poor trend following instrument and it would be tempting based on past performance to eliminate it from the portfolio but alas to be fair, and not to curve fit, I need to retain it in the portfolio. Hopefully sooner or later it will have it's day in the sun with trend following. Have a good trading week mate :-)

Parking D13 and B1 Composite Strategy Temporarily

As we are now getting to the pointy end of this composite strategy and need additional equity curves of particular discreet profiles to fill the drawdown gaps, I will be temporarily putting this composite on hold while I conduct further testing of alternative systems across the portfolio. Never fear....we will return to it when we find a suitable set of equity curves to find the balance of unexploited market conditions during this period. If we don't find any, then it is on to the next composite test and so on and so forth.

We will note that this composite has been temporarily parked for further investigation so we can return at a later date when or if we find another suitable system for inclusion.

The status of the D13/B1 Composite is as follows. Below is the drawdown profiles of the individual systems and their composite.



The yellow highlighted zone is the problematic period for the composite strategy where positive drawdown contributions from both the D13 trend following system and the B1 mean reversion system are exacerbating the composite system's drawdown. If we can reduce this drawdown to a sufficient level we will then be able to further scale up the composite strategy to hopefully meet or exceed the TF Index performance. The good news is that the two systems outside this period smooth the overall drawdown of the composite system so all is not lost.

Just for my own reference, the problematic instruments during this period that contributed to the drawdown were as follows:
EURAUD;
EURCAD;
USDJPY;
EURO-BUND;
BRENT OIL;
HEATING OIL;
NATURAL GAS;
WHEAT.

Attached is a spreadsheet of the blended results of D13 and B1 across the portfolio over this problematic period for future reference in identifying broad market conditions that could assist in strategy development.

Things slow down at this point as it is necessary to interrogate the poorly performing instruments during this period 1/1/2002 to 31/1/2006 and specifically identify the general market conditions during this period to hone a suitable strategy that can offset the balance of the other contributing system's drawdowns.

If necessary we can resort to this troubleshooting process.....but this can occur at a later date......as I plan to steam roll on in developing alternative systems to build a database of equity curves which we can then use at will in our composite system strategy development. If we find a suitable equity curve to plug the gap, then we can return to this system to improve it's performance.......until then it is onward and upward.

Cheers

C

Did he mentioned what these systems are because there is some nice equity curves in there.

No Bass.....he was fairly non-committal regarding any systems he used to generate those equity curves.....hence why I am having a crack myself at getting familiar with the technique so hopefully I can start generating similar results. I am starting with the simple systems as there is less chance that they are curve fitted........I am thinking that I might replicate the trend following systems from Wisdom and see how those systems fare against my particular universe.

I will post results on this thread as I do it so we can discuss them. As I plough through all this Bass I would appreciate your ideas and comments along the way.

C

I'm doing this too C., i'm working day and night to improve my systems with the little time that i have free, as i mentioned several times before this is not an easy task and so is making consistent profits in the Markets.

We will get there mate...and when we do....it's hasta la vista to the day job baby :-)