Systematic Trend Trading

Portfolio Compilation Step 1 - The Raw Portfolio

Let us say that you have a number of different strategies and/or systems, each with varying volatile equity curves. The temptation is to select the overall best performer and go with that one, but this method outlines a process where you use all of them and then risk treat them to compile a portfolio that achieves a far great result than the sum of it's parts.

My other thread looks at a semi-automatic momentum breakout system on M30 which has been tested across numerous markets (24) over the last 10 years. I will use the results from this system as a basis for explaining how to compile robust portfolios. About 19 of the 24 tested systems have what I regard as a slight edge. Over the 10 year horizon they simply managed to be profitable. While there is no guarantee of an edge....there is a good chance that an edge is present in these 19 systems given the trade sample size for each and more importantly the range of different market conditions that have been experienced by each.

I use all of these slightly profitable systems for my portfolio compilation and avoid the temptation to cherry pick the best performers. This does not mean that the remaining poor performing 5 markets are not continuously tracked and backtested. They are. If they become more profitable over the longer term into the future then they will also be included in the portfolio at a later date. We just do not include them now based on their track record.

So below is a chart of the individual equity curves over the 10 year period of each slightly profitable system.



....and this is how they look when superimposed on the same chart for a spaghetti experience.



Bloody awful aren't they? Some are ok....but many are all over the shop in terms of overall volatility.

Here is a table of performance metrics of each strategy.



The temptation would be to select XAUUSD from the list given it's overall result (refer to yellow highlight). This produced the best overall return (a CAGR of 13.7%).....however if you look at the MAR ratio column, in risk weighted terms you may have found a better choice using MAR. With different max drawdowns such as USDCAD (highlighted in pink), the difference in apparent profitability is simply a factor of volatility leverage.

For example Gold has a max draw of 22.49% and a CAGR of 13.7%. USDCAD has a max draw of 9.65% and a CAGR of 8.1%. I could double the position size of USDCAD to achieve the same max draw of Gold at 13.7% and in doing so, my CAGR would be approximately the same %.

Now look at the result of simply adding these individual systems together assuming a total $50K portfolio.



Firstly note how simple addition has reduced the overall drawdown to 8.74% for the entire collection. This is attributed to the fact that drawdowns of each different system occur in diferent places.

Now also note the MAR ratio of 1.77 which is far higher than what can be achieved at an individual strategy level. This MAR ratio means that pound for pound, the portfolio is about two times more powerful with an equal allocation of capital than your best single system performer (XAUUSD). Now remember we have not applied selection bias here. We have used all those shitty return streams with a slight positive edge....and not simply the best one.

What we are doing here with diversification is optimising our risk results. We are not optimising our profit results. Risk and return are inherent in any asset. We can manage future risk but without a crystal ball we can do nothing about future profits.

Diversification's free lunch resides in the following principle: The impact of addition suppresses the individual drawdowns of each return stream and the positive edge in each system is magnified by amplification. Even from simple raw addition, without any further portfolio treatment, your risk adjusted returns are significantly improved. This allows you to scale up global position sizing across all systems to reach for the stars in terms of CAGR while still managing your drawdown.

That is all good in theory with a backtest, but how do you manage this in practice in the live environment? Well you commence with the backtested 19 slightly profitable systems and continue to monitor their profit factor. As soon as the backtest plus walk forward result dips into negative profit factor territory, you turn them off. You still continue to monitor them...and when they slip back into an overall positive profit factor....you turn them back on. Over time markets will slip into and out of your live portfolio based on this principle. You are not looking at highly profitable systems here. You are just including those systems in your portfolio with a slight positive edge....no matter how weak. Later we will be utilising these curves to do some wicked things at the portfolio level.

Next post will look at Step 2 Portfolio treatment which standardises the volatility of each return stream in the portolio.

Cheers

C

Portfolio Compilation Step 2 - Risk Adjusting

This step taken in this risk treatment phase refers to the standardization of volatility across all return streams.



Refer to Column C. This is the original drawdown of the different strategies from the prior post when untreated.

We first decide a suitable max drawdown for each system in the portfolio. In this case in column D I have chosen a desired 10% max draw for each system. Generally think in these terms. If you have a risk appetite of say 20% as a max draw as general guideline for the entire portfolio, then half the max draw of each individual system comprising it. This ensures that your ingredients are not too highly leveraged and will produce a result at the global portfolio level that is well within your risk appetite tolerance.

So in column D we use a 10% max draw per system. You then use this as a basis to determine a multiplication factor that is applied to each raw system result. For example AUS200 had a raw max draw of 4.4%. If we wanted a max draw of 10% then you apply a multiplier to your position sizing method of that system of 2.27 x.

Having applied this multiplier to different systems we can then refer to column AC of Max Draw and see that each system is now producing a max draw of approximately 10% subject to lot sizing rounding and minimum thresholds.

Now have a look at the CAGR of each system and notice how it is re-ranked from how it appeared in the raw portfolio. You will see that USDCAD and XAUUSD now have similar CAGR for the same approximate risk exposure (where risk is volatility per unit time).

Now that we have risk adjusted the entire portfolio let's see what the total result on a $50K portfolio produces.

Before



After

Attached Image (click to enlarge)

If you compare and contrast the MAR ratio of both the before and after case you will see that the 'after' case has a better MAR ratio. This tells you that pound for pound, the risk adjusted portfolio is now more efficient and powerful. Don't be distracted by the overall lower CAGR of the 'after case'. When you have a higher MAR ratio you can then scale the 'after' result to achieve the same desired drawdown level as the 'before case' and you will find that the Risk Adjusted portfolio produces a far higher CAGR with the same level of drawdown.

My next post will look at the third stage of portfolio compilation where we look for weaknesses in the system compilation that can detract from overall portfolio performance and eliminate them.

Cheers

C

Portfolio Compilation Step 3 - Eliminating Return Stream Weaknesses

This is an essential part of portfolio construction. The aim of this step is to identify any return streams with weaknesses that could compromise the overall global portfolio.



They key factors to look at in this phase is to :

1. Eliminate strategies with negative skew - If some of your strategies have negative skew (eg. many small wins with occasional large losses) then you want to eliminate them from the portfolio. The consequences of a sequence of large losses if they arise can seriously effect overall portfolio performance. Remember the portfolio works on the principle of 'the sum is greater than it's parts'. A single return stream that dominates the overall result compromises that intent. What many don't realise is that diversification is only ever effective on positively skewed strategies. Negatively skewed strategies tend to be predictive in nature and require concentration as opposed to diversification. In this particular example, because we are using a momentum breakout system which cuts losses short and lets profits run....we don't find any examples with negative skew (refer to column S). They all have great positive skew;

2. Eliminate strategies with negative MAR - This eliminates those return streams with a profit factor less than 1. We have already discussed this in the first post...so no need to go over it again. We have already filtered out all strategies which do not have a long term edge; and

3. Eliminate strategies that are broadly too highly correlated. Refer to column R. WE are looking for any strategies during this broad brush process with correlations approaching +1 or -ve 1. In this case there are no strong candidates but I have highlighted those that might be a bit problematic but we will use visual methods in the next step to eliminate them if necessary.

Now that we have eliminated any portfolio weaknesses and have a very solid robust suite of systems in the collection we can now proceed to the final phase (Step 4) where we map the equity curves and really up the ante to blow away the competition.

Cheers

C

Portfolio Compilation Step 4 - Mapping Correlations to Reduce Global Portfolio Drawdowns

This next step is an essential component of portfolio compilation that many....just don't get as they like to deal with single statistics. Hopefully I can explain it in a way that makes the light-bulbs flicker. This is probably the most important step of the entire process that literally magnifies your performance returns at the portfolio level several fold.

Remember that we like correlation when markets are trending but avoid correlation when markets are not trending. Using a single correlation statistic throws some great combinations out. That's why mapping is a far better method for identifying adverse correlation impacts. Only focus on the drawdown impact.

So here is the visual result we are currently left with after step 3 of this exercise. The equity curve that is repeated across the top is the overall portfolio result. Below lies the individual risk adjusted return streams after stage 3 treatment. The reason why I repeat the portfolio curve at the top is to assist me in identifying which systems are contributing to unfavourable drawdowns by mapping.



If I closely look at the portfolio result below, I can see where the biggest drawdowns occur. In this case around April to August 2010 and also April to 2018 to April 2019. The reason for this drawdown profile during these times is attributed to correlation that exists between systems at the individual return stream level. You need to find these culprits and where possible eliminate the impact to significantly improve the overall result and make it more robust.



The way we find the culprits and eliminate them is to map the equity curves against the portfolio. I use vertical red lines placed over the max drawdowns to see what is going on. As I turn of different combinations I always refer to the MAR of the adjusted composite to see if there has been an improvement or not.



Now I have found a few culprits and have turned them off. Notice that I don't have favorites in this selection process. My total focus is on the global portfolio result. The systems are just a means to an end. It is the portfolio and how it is compiled that is most important.



Now because this is an example I will stop there.....but you will notice in the chart of remaining systems above how we still have drawdowns (but far lower) in the portfolio. I could find the culprits this way and turn them off to vastly improve the result again....but it is a long winded and tiring process....so I will stop here just to demonstrate the power of this step.

Lets now see the before and after result.

Before



After


Look at the equity curves of before and after and look at the MAR metric.

You will notice that the portfolio equity curve has been flattened and the MAR has increased from 1.91 to 2.35. It is hard to see the impact on overall performance but the following will definitely demonstrate the impact.

We allocated $50K to the Risk weighted portfolio (before case) and achieved a CAGR of 11.37% with a Max Draw of 5.94% but now we can allocate a lesser sum of $35,000 to the compiled portfolio (after case) to achieve a similar Max Draw of 5.82% but a higher CAGR of 12.59%.

Here is the result on the compiled portfolio with a $35K allocation.



Now many might not think this process is big biccies....but let's summarize where we started from and how we got here in financial terms.

As a raw portfolio if we had selected the single best system only (XAUUSD) as the system to progress a $35K allocation towards it would have resulted in a CAGR of 8.4% and a max draw of 10.39%. If we had applied aggressive position sizing to squeeze out the juice to a desired 20% max draw we might have produced at best a result of say 16%.

The raw portfolio result as a compilation on a $35K allocation produces a CAGR of 18.5% with a max draw of 11.9%. A slightly better result than picking the winner.

The risk weighted result on a $35K allocation produces a CAGR of 14.0% with a max draw of 8.3%. An even better result than the raw portfolio as I can scale this up to achieve the same desired drawdown and achieve a higher CAGR.

The final compiled version on a $35K allocation produces a CAGR of 12.59% on a 5.82% max draw. A far better result than all options which is scaleable and can deliver far greater bang for buck on your finite capital than any alternative.

For instance with a 20% max draw tolerance I allocate $8,500 towards it and achieve a CAGR of 24.9%. Kaboooooooom Tiiiiiiiiiiish

This is all through addressing the risks through portfolio management and is fairly independent from the individual systems comprising it. Can you now see why so many FM's operate a diversified portfolio? Where are the returns coming from.....the systems or the portfolio treatment? Imagine if your inputs were uncorrelated powerful risk adjusted systems The mind boggles. There is more to it than simple compilation, but when you get the hang of it....you can start competing with the pro's.

This is more like a chef than a quant guys. You start with your quality ingredients (with a slight edge) but what makes the meal is in how it all marries together.

Cheers

C

Signing off for a while. That was exhausting.

You guys (Copernicus and 60 minuteman) makes my level of dopamine constantly high …
You write those precious texts faster then I am able to absorb them …
Thank You Very Much …

thank's Blacknapkins!

and thanks C, there a lot of info there to digest, I'm going to work through it stage by stage and see if I can build a small portfolio here. Inspiring stuff and massive respect for putting in that effort.

Week 50 -9R

Results: 434 trades, +129.5R

ALL LOSERS THIS WEEK WITH 2 TRADES STILL RUNNING - Initially it is hard to get used to runs of losers, but with repetition their impact fades and your mental clock adjusts to the longer term horizon. Much of the work is done by constantly repeating the same process, reading inspiring posts can spark the imagination but truly learning a skill, and it is a skill, comes from boring mechanical processes.

Two weeks left on this test

Ni wuzzas 60. You and a few others can peer through the apparent text complexity and see the secret sauce. I get traction with the FM's with this process but in the retail world it is a bit like this.



If you do this yourself you will find many surprises along the way. It is far better to take this and develop your own process than me simply providing a few spreadsheets. The guiding tips in the above posts should prevent roadblocks along the way.

What you will soon realise is that if we banded together with our systems that may possess a small edge, then collaboratively we could put together some really powerful beasts that would be the talk of the town :-)

A big thank you to Darrin and Copernicus for these absolutely eye opening posts regarding asymmetrical reward/risk trading and the power of portfolio systems management... words cannot adequately describe the impact that these concepts will have on my trading...

best,

Dave

On my third read of this...brilliant stuff and feel inspired to give my random jumble of systems the portfolio treatment... first job, data!

Has anyone considered the use of a dynamic profit target as opposed to a trailing stop.

If the Trailing stop was seen as a worst case scenario when the trend suddenly changes direction, a dynamic exit should result in less left on the table.

The question then is what form does this exit take?

Ideally it should give an early warning of a change in trend. if caught early enough, but not too early, your returns should improve.

any thoughts?

Hi RLF...

It's an interesting concept, here's my initial thoughts:

1. A trailing stop works well enough, not only in my testing but also the extensive history of trend following returns
2. A dynamic exit would require some sort of prediction of future prices based on past prices - I wouldn't like to say it couldn't be done but I am yet to find any significant evidence it can be done.
3. If you try and measure trend changes with any accuracy you become susceptible to reacting to noise - If you use a 200sma you will be late to the party but wrong less often, if you add some accuracy to the timing, such as a 10sma you are early to the party but generate more false signals - I believe it is called the uncertainty principle in physics - the more accurately you try to measure the momentum of an electron/neutron/particle (something like that) the more off you get on it's location (it's something like that, would need to wikipedia it), but it's the same principle. There is always a trade off between timing and accuracy/false signals. What trend traders (or price followers) do is forego 'best' for 'good enough'

here you go...

The uncertainty principle states that the more precisely you measure one quantity, the less precisely you can know another associated quantity. The quantities sometimes come in set pairs that can’t both be completely measured.
What Heisenberg found was that the observation of a system in quantum mechanics disturbs the system enough that you can’t know everything about the system. The more precisely you measure the position of a particle, for example, the less it’s possible to precisely measure the particle’s momentum.

Jerry Parker and the TTU mob were talking about this on their podcast and referring to it as the bane of their existence....leaving too much profit on the table when the trend ends. Neils suggested that Dunn had invested a lot of time in looking at ways to achieve this at the portfolio level...suggesting that perhaps there might be a method. Not sure exactly what he was thinking.....but in thinking about it....perhaps you could deploy short term, medium term and long term models (all with an edge over the long term) but exhibiting different behaviours during massive trends eg. exponential accelerations. The short term would more closely hug the exponential move etc. etc. etc. so the end result of the combination might be better than what could be achieved on a single system....however when markets were not trending or trending slowly (not exponentially) then the short term and medium term would most likely suffer more whipsaws (as the uncertainty principle suggests) compromising the better profits arising from when the trends soar.

I have been thinking however that maybe....you could deploy filters based on the performance of the long range system that only allowed the short term and medium term solutions to only be active during exponential moves on the long term. Turn them off in all other instances. You might set the short term with a 2ATR trail, the medium term with a 3ATR trail and the long term with a 4ATR trail. This might allow you to squeeze more juice out of the 4ATR trend before it ends. It probably wouldn't work as my track record in this blubbering thoughts always has a sting in the tail....but it might be worth testing.

This is another simple strategy I trade manually - I would love to get it coded, so if anyone is game?

Instrument GBP/USD
timeframe H4

1. direction filter 200sma
2. price hits 20sma
3. price closes under 20sma - sell at close - stop at local high
4. trail exit
5. If already in a trade take all additional signals as long as they meet this rule - if already short additional shorts should be at a lower price (see entry 2, entry 3 - both valid)

That's it, nice and simple, theory is that forex markets display more volatility in the short term causing prices to revert to mean (the move back to the 20sma) before continuing with long term trends (the 200sma)

Don't get me started on Quantum Mechanics 60.....Stop teasing me :-)

Nice one 60. Where is a good coder when you need them?....*eyes wandering back a few posts/tic* :-)

Someone mentioned it on Jerry Parkers twitter... I thought ooh that sounds right up my street (apart from the physics bit of course)

Pffft....physicists....what do they know. Don't let them get in the way of a good chef. You can't eat a Schrodinger equation. It is full of dead cats. :-)