Diversified Trend Trading Approach

I am not sure your stats offer much to dispel the idea of entering early. As i understand it your entries are based on breakouts, or higher highs, lower lowers once a trend identifies itself by producing a valid SDC.
As such your results are on trades entered late in my opinion.
If the SDC is valid, it should produce similar results to those you posted, regardless of when you enter a new trade. So why not enter earlier and possibly push more results into the 2R+ regime.

The only way to drill down into this assumption is to backtest the discretionary method over the same extent using shorter trend entires on average...any takers

Do entries matter?

Most of us are aware that it is expectancy that determines whether a trading system is going to be profitable or not as opposed to merely the system win rate. What does this mean?

Expectancy = (Pwin x Average Win) - (Ploss x Average Loss). ........Alternatively you can simply calculate total profit/loss over a total sample range and recover the same result by dividing the total profit/loss result by the number of trades.

There are two components to expectancy which are critical in understanding where the secret sauce in trading lies. The first is the Win Rate which is the probability of a winning trade and the second (often disregarded element) is the Average profit per trade.

Now trading results differ to a simple coin toss experiment by virtue of the fact that in a coin toss, the average profit per toss can be excluded from the result as a win is the exact same value as a loss. For the example below assume a win is $1 and a loss is $1 . The result is that the result of the coin toss experiment is solely dictated by the Win Rate. No matter how long the coin tosses in the air, the length of time of this event does not interfere with the Expectancy of a coin toss. This differs to trading as average profit per trade is a variable that is dependent on market conditions in the context of how the constraints of your system deal with those market conditions. For the mathematicians out there you actually need to break up the entire trade into discreet time intervals and then determine the $ expectancy of each interval and through summation through integration, arrive at the total expectancy result. The central message of this is that the price action over time of the trade from entry to exit matters to the overall result which is not evident in a coin toss experiment.

Effectively we are referring to the methods of trade management applied over the course of individual events within the trade itself. If your system constraints possess sufficient degrees of freedom to allow for the market to breath, and rules that hone the trade management to a particular directional market bias, then if market conditions favour that regime and overall bias, your average profit per trade improves. This therefore allows you to bias your system away from simply entry decisions that apply at a discreet time interval (the time of entry itself) and establish a directional bias in an extended time interval that has more direct relevance to the overall result (the price action that occurs over the entire time interval between entry and exit) and furthermore is directly in your control to manage.

In general all trend trading systems apply this principle in the following mantra - "Cut losses short and let profit run". = This can be more definitively stated by using a trailing stop a sufficient distance behind volatility and choose a directional bias in direction. This principle focuses performance efforts towards the average profit per trade as opposed to expending effort on Win Rate. It is recognising that over the entire course of a trade, an overall bias will tilt probability in your favour as opposed to attempting to define a specific point in time as a probabilistic measure of success.

Coin Toss Expectancy = (Pwin x Average win) - (Ploss x Average loss) = (0.5 x $1) - (0.5 x $1) = $0.00 per toss on average. Multiply this by the number of tosses and estimated return is $0.

This differs from real trading as the Average profit per trade is dependent on the nature of the market over the course of a single trade. For example if the market is truly random then overall expectancy will be $0 (excluding transaction costs and slippage). This is because both the Win Rate AND Average Profit per Trade for a truly random market is 0.

If however the market is not truly random and there are periods of time where non-random uni-directional extended price moves exist, then either the Win Rate OR the Average Profit per trade can bias the total expectancy result in your favour. You can determine which element of your strategy is responsible for an overall positive expectancy result by running sensitivity tests on your strategy that drill down into the strategies key variables such as the overall result if trade entry is random versus the overall result if trade entry is non-random. You will find that their is very little edge delivered by the entry itself when applied to trend trading. This is really a no-brainer to someone that understands trends as it is the price action along a defined time interval that defines a trend and not merely the condition of entry itself that relates to a discreet time interval.

Tom Basso and Van Tharpe highlighted this principle with their random entry trend trading system whereby a purely random entry system applied to a sound trade management approach designed to capture non-random market action resulted in a profitable system across 10 liquid instruments for a period of 20 years producing an average win rate of approximately only 36% or so. While market conditions have changed over the past 10 years or so for the trend followers out there, the system performance of this random entry system possessed a powerful underlying message. That the dominant contributor to performance for a trend follower relates to trade management as opposed to entry selection.

It does not mean that you should totally disregard entries, but just means that you should not be obsessively focused on this aspect as most retail traders are.

Once you get a handle on this then you do not need to be able to predict market behavior but just recognise that over an extended timeframe, there will be periods of non-random directional price behaviour. By applying a systematic strategy that rigorously manages risk to keep you alive long enough to capitalise on non-random market behaviour but has a directional bias is all you really need to gain an edge.

To give you peace of mind you just need to understand that at times there is uni-directional behaviour expressed by it's participants and this is perhaps best explained by market crisis events where collective flight or fight tendencies prevail or alternatively recognise that underlying economic fundamentals will over the long term reveal themselves as underlying bias in otherwise random market behaviour.

yes, they do

Hehehehe. It all depends on context and the system you use. The above is flavoured towards the trend folllowers and for the long term. If you can achieve a a positive expectancy over the long term and entries are materially significant to your system, then I am with you......but notice the 'IF' in this context.

Well said.

Way too much emphasis on entries. I guess it possibly stems down to the fact that R:R is shoved down everyone's throats with the concentration on 'perfect' entries and small/tight stops yielding greater overall performance? There seems to be little focus on the fundamental reasons WHY to enter and even less focus on fundamental reasons to stay in or get out of a trade.

For the pure technical traders - I wonder what the effect would be if people couldn't zoom in on their charts (therefore couldn't see the candlesticks) and only had the daily and above to use?

Like below.

Agree entirely.

What most of us retail traders forget about is the enduring impact that other participants have upon the price of an instrument over the course of time. Even with effective random entries taken at the retail trading level from trading short term jitters, the effect of overall participant bias from fundamental investment behaviour or long term hedge behaviour of participants will progressively dilute the random noise on the shorter term timeframe.

To avoid the noise and make a sufficient return you need to reduce your return expectations on a single instrument, set wide trailing stops well outside short term volatility and trade a diversified portfolio (while keeping your fingers crossed that unidirectional extended events will occur) ....unless you are sufficiently confident with your fundamental analysis on a narrow range of instruments. I can't see any other way out of this dilemma.

definitely. if you dig deeper, there is always something down there.
however if your initial position means something, the entry definitely matters. for example, when i speculate the reversal of a trend when exchange rate's general direction is upward, my initial entry is low leverage. and then i keep building positions if i was right with my assumption and maximize the profit. if was wrong i bail out. when we talk about a system where entry does not matter, the efficiency/profitability of trades falls significantly. this is a topic for discussions if trading is worth at all. i would not be involved in something that does not pay for my efforts, time and investment with leverage. therefore my conclusion is, if i am investing money and my time, they must pay off in maximum. otherwise i would go to mcdonald's as part time stuff, this would be more efficient profitwise, rather than getting involved in trading where entry does not matter

I hear you Tash and totally understand your perspective. I can only speak from my own experience. I do not rule out the potential of what you say and continue to look at different systems, some with defined entry conditions....but have yet to find that they materially contribute to overall success....however mean reversion systems seem to hold some weight in say the last 10 years...... but I am not so sure this is sufficiently robust to handle the long term and am suspicious that the last 10 years are not 'normal times' for the market given Central Bank intervention practices and Quantitative Easing programs.....plus with the increase in bots over this period and HFT. It all depends on what we define as 'robust' and in accordance with our risk tolerance for our level of trading capital. I hate losing.......and therefore have a defined risk tolerance which is aligned to the very long term.

yes. every method/system has its own negative and positive sides. long term trading has to deal with black swan events which can wipe out significant profit gained for many months. and then you have to identify if trend is reversing or just bouncing to secure the profits. the other issue is ranging markets for months sometimes for years which significantly make long term position trading ineffective. if you scalp you have to deal with high volitality and spreads etc. the same issues apply to intraday and mid term trading. therefore everything depends on one's expectations

Actually it appears that it is the crisis event, where long term trend trading shines. There will be 'black swan events' where slippage will eat into trend performance but there will also be 'white swan events' for a trend trader where profits are magnified and without a profit target, will deliver alpha. Unlike most other trading systems, trend following capitalises on market divergence as opposed to market convergence. Drawdowns are delivered through long consecutive strings of small losses as opposed to single major losses like other 'convergent systems'. If you are interested have a read of this research paper which is an extract from her great book based on results of 800 years of market data. Without the crisis events, medium to long term trend trading would be forced to achieve beta of say most short-term debt instruments. The risk of course is the associated drawdown between these major events and is accentuated by over-trading during these periods. Using supplemental systems such as mean reversion systems therefore make sense if you are seeking to reduce volatility of returns and achieve better cashflow pushing your combined system towards the most efficient performance metrics.

Thoughts on Automated Systems and System Diversification

While the subject of this thread is on a discretionary trend trading technique, I have had a history of involvement with EA developers looking at fully automated systems and for the past 12 months have been spending some more time testing systems across a diversified portfolio over long term data sets.

The following is an opinion only from direct experience in crunching the data outputs of these systems, but I am becoming more confident in the assertion that the reason for the noticeable absence of long term profitably EA's in the market, is that unlike discretionary trading styles, an EA is typically programmed to rigidly respond to a specific set of market conditions and as a result has a limited lifespan as market conditions change. Many on this forum would be fully aware that it is near impossible to find an EA that is profitable over an extended timeframe beyond say 5 years. I know that I am not alone in this conclusion.

Despite this limitation, the benefits of fully automated systems still holds significant appeal so is there any way out of this dilemma.......and the answer is a resounding yes with a few caveats.

The broad principle surrounding this discussion is how to consolidate specific EA's that respond to a given set of market conditions into a portfolio of systems.

The broad categories of market type can be broadly categorized according to the following table.



The aim therefore is to develop EA's that specifically respond to each market condition and are sufficiently profitable over the long term (say 10-20 Years) to at least preserve capital or profitably contribute to capital accumulation. In fact you can have some unprofitable systems (within tolerable thresholds) that are effectively subsidised by the performing systems to achieve a flatter equity curve which is the intent of this discussion.

For instance a simple trend trading EA such as a 200 period moving average crossover system is likely to perform well over two of the 6 market conditions (quiet up and quiet down), whereas a momentum based breakout EA is likely to perform well in (volatile up and volatile down). On the other hand a Bollinger mean reversion system for example is likely to perform well during both volatile and quiet sideways markets.

Now the reality is that you are going to find over the long term that it is unlikely that you will be able find an EA on its own that will produce CAGR of beyond 10% without unduly entering high risk drawdown territory. For example, over the past 20 years, of all the simple systems I have tested the best result in isolation for any system has been approximately a 10% CAGR with a 20% drawdown. I am finding as a rule of thumb you could assume that you can double your average annual return to determine the likely level of drawdown to expect. What limits the ability to enhance your returns is your volatility of returns and the likely level of drawdown you will experience over the long term. For example I could simply double my risk% per trade and double my return to say 20% provided I am prepared to tolerate a 40% drawdown which is well beyond my risk tolerance.

.,......but what happens to that maximum drawdown when I combine these different systems. Even if I assume that each system generates an average return (CAGR) over the past 20 or so years of 5% per annum with an associated 10% drawdown, when I combine these systems I find that that I can achieve a 5% CAGR with only a 3% drawdown. Why?.......because each system contributes to the equity curve when the other systems are faltering and entering a drawdown period. This is the theory behind why you should consider systems diversification as a method to respond to changing market conditions.

What this means for a portfolio comprising diversified systems is that if I am prepared to tolerate a max drawdown of say 20% I can multiply my trade risk % per trade by 20/3 = 6.66 and therefore multiply my CAGR from 5% to (5 x 6.66) = 33% return.

Now remember this is just a theoretical example and an analysis of past data history is no guarantee of future success, but hopefully it shines a bit of light on some simple techniques you could consider applying to improve your performance returns without having to expose yourself to unnecessary risk.

Have a great weekend

C

Here is some great listening material for the week ahead from Rob Carver on 'Top Traders Unplugged' speaking about simple automated systems and diversification.

Making a Simple System and Sticking To It | Robert Carver

Why Diversification is Key | Robert Carver

Interesting Ja

For what it's worth the systems described above in my post are actually turned on all the time as I simply can't predict with reliability when market conditions change. The key therefore for consolidating systems is that each on it's own merit needs to either be profitable over the long term under all market conditions, or at least a marginal loss maker that is subsidised by other profitable elements.

What has really been bugging me is the degree to which my trading capital is tied up with a single trading system as I always need to plan for the worst with a drawdown. The composite system approach seems at least a way to overcome this very unproductive use of capital to some extent without being overly risk exposed on the portfolio.

With regards to your analysis of trends, at least within the trend itself you could employ two methods to harvest opportunities. For example you may decide it is worthwhile catching swings AND the long trend trend itself by releasing say two positions on entry of which one position or a portion of the position is allocated towards a swing with a tight trailing stop with the other position left with a large trailing stop to catch the longer term trend move. Alternatively you may decide to delay entry of the swing component after the trend has matured and you have at least exceeded breakeven on your original position to avoid excessive correlation impacts. By employing both components you at least capture some of that glorious unrealised profit that you typically lose with the longer term trend trading system on exit. Furthermore, the composite nature of two different systems capitalising on trends is also going to assist in reducing overall drawdowns.

Lol... your too bloody quick !!!


I removed the post thinking sometimes I ramble on a bit too much.. sorry mate.. lol..

No probs mate...and it wasn't rambling....... I am Speedy Gonzales tonight

Lol...

I do this already by separating the Monthly and Weekly charts into signal charts . if we then step down a chart for entry and trade management this creates the different time lines for exposure..

Nice mate. That will do the trick.

Systems Diversification to Improve Performance - An Example

I have spent this morning compiling results of 2 discreet automated Trading systems to demonstrate the logic surrounding post 512 of this thread. The purpose of this explanation is not to provide a profitable composite system in itself as I will leave that work to those who may be interested in applying these principles. Rather the intent of this explanation is to demonstrate how to consolidate algorithmic systems into a collective basket to improve system performance efficiency. I will therefore only delve into melding two (maybe three) disparate systems to demonstrate what diversification of systems brings to overall performance.

Also remember that to get a robust system, 6 years of D1 data is simply not enough to cover a sufficient sample that is representative of all market conditions. I have for the purposes of this demonstration only used about 6 years of data, but for more rigurous testing would be looking at between 20-30 years worth of data.

Each system has been tested across a medium term data set commencing 1 Jan 2000 to 30 Nov 2015 and across a portfolio of 46 instruments spanning across asset classes (the instrument list is at the bottom of this post). Note that no science has been applied to the selection of instruments such as equal weighting of instruments by asset class which would be a wise move. However for the purposes of this exercise the number of instruments spanning low correlated asset classes is sufficient to achieve the result I am seeking to demonstrate.

System 1 (summary results and equity curve below) is a very simple Long Term trend following system that is designed to capture nice slow low volatile short/long trends. Long term trend following systems in general have performed poorly between 2010 to 2013 (backed up by CTA performance over this period) but have experienced positive performance between 2014 and 2015 chiefly attributed to more volatile market conditions across asset classes. The broad reasons for this increased volatility during this period across asset classes can only be guessed, but more than likely this can be attributed to China's economic volatility. What the tests on this long term trend following strategy reveal is that despite the good times of 2014 (8%) and the great results for 2015 of 34%, overall returns (CAGR) for the 6 year period have been paltry at 4% PA. This level of return has been scaled according to my total risk tolerance profile of which a max drawdown of approximately 20% is the most I am prepared to tolerate. Of course for the bold amongst us if you simply doubled your average position sizes, your return profiles would also have approximately doubled but with it comes the cost of an approximate 40% drawdown.



System 2 (summary results and equity curve below) is a very simple Short Term Momentum Breakout system that seeks to capitalise on volatile market conditions for both long/short conditions. The strategy simply held it's ground between 2010 to 2013 but experienced strong performance in late 2014 (20.88%) and a nice recent kick in performance in late 2015. Overall CAGR has been barely above breakeven of about 2% with a scaled max drawdown in mid 2014 of 20%.



Pretty hopeless results for each system in actuality.....but never mind...... as I just want to demonstrate what happens when we merge both systems as this is where you leverage performance by diversification.

Before we get into merging the systems, pay close attention to the equity curves and the months where significant drawdowns occur. For example the peak drawdown in System 1 occurred in Mar 2013 with another nasty drawdown period in May 2011. For System 2, the peak drawdown was experienced in May 2014. You can see that the merging of both systems should have the following effect. The peak drawdown of System 2 will be reduced by the equity contribution of System 1 during this period.

If you are familiar with the effective constructive and destructive interference in wave patterns in physics, then visually you can start to apply this effect when merging equity curves. What you are looking for are ways to reduce the ugly drawdowns by integrating systems which perform during those drawdown periods. You are therefore by definition integrating systems that respond to different market conditions in ways that achieve overall superior performance with your limited capital. What this practice in essence is doing is identifying bundles of non-correlated systems that respond to different market conditions.

Ok...so here we go. By merging the systems the following performance metrics are achieved.



Note the generated equity profile and how the peak drawdown is now shifted to a new location (being May 2013). Now while this also has been scaled to achieve a max drawdown of 20%, note the Adjusted CAGR column. You will see that we now have strong performance in both 2014 and 2015 months of >24% each month and an overall improved CAGR for the 6 years of 5% PA. This is a 25% increase in profitability in the combined system against the best performing individual system of 4% but with no increased portfolio risk in terms of maximum drawdown experienced.

......now while this was a quick and nasty demonstration and you all will be shuddering in the meagre performance of these systems over the period, what I am demonstrating here is that by system consolidation you can actually manage your risk return profile. The ultimate aim is to use this method to cumulatively manage your portfolio equity curve and the principles here apply to any method of diversification whether by diversifying by market (instruments), diversifying by timeframes traded or by diversifying your systems (as this example demonstrates). To give some hope to some interested traders out there it is possible through this technique to achieve sustainable returns of between 20-30% PA over a 20-30 year period within acceptable drawdown tolerances of between 20-30%. The key is how you construct your portfolio to achieve this.

With the example above you can see that we have now incorporated two techniques that respond to trends of different nature. Obviously to improve overall performance of the portfolio the next logical addition to improve overall performance and iron out the wrinkles in the equity curve would be a system that traded sideways markets. You then reexamine your equity profile and identify the weak spots and go again with a system that capitalises on those new weak spots.....and so on and so forth until you achieve your desired return and risk tolerance. By using extensive data history the chances are that you have developed a range of systems that can respond to a range of market conditions without having to predict when to turn which system off or on and when.

Most of the large CTA's out there have greater than 10 systems running at a time. If each system say for a 20-30 year period across a vast range of market conditions can contribute between 2-3% of CAGR, then you have a system that can generate significant CAGR (eg. 20-30% PA) within a defined tolerated drawdown of say 20-30%.

Rather than spending the effort seeking the holy grail in your EA's simply re-direct those efforts to examining how you can collectively diversify your portfolio to optimise your risk reward profile. What you will find is that you end up with a composite system that within strict risk management guidelines can respond to a variety of different market conditions. This is where you can then turn to the successful discretionary trader out there and say....move over, it is time for me to invade your space.

Systems Diversification to Improve Performance - An Example - A Continuation

Continuing on with the example in the previous post I have now had a chance to test a simple CounterTrend EA over the instruments and data sample and have included the performance results below.



As suspected the market conditions for the portfolio between 2010 and 2012 clearly was predominantly in sideways congestion, as the performance of a short term CounterTrend strategy was quite strong over this period contributing CAGR of between 14.26%, 19.23% to 18.47% for 2010, 2011 and 2012. Yeah.....this strong performance over this period is going to significantly contribute to flattening the equity curve of the combined systems.

During the years 2013 to 2015 where trend following started to pick up performance the results of the CounterTrend Strategy deteriorated producing a maximum scaled drawdown of 20% in March 2015 at approximately the same time that the trend following strategies were shining. Despite the poor performance in 2014 and 2015, overall CAGR for the 6 year period was 9.6% which is a pretty solid result. The benefit of analysing market conditions this way is that you get a feel for what systems work best in what market conditions.....but don't get too cocky. You are only working with 6 years of data in this example and for a robust strategy you need to cover as wide a range of market conditions as possible....so think in terms of 20-30 years before you get the grin of confidence.

However........back to optimism........clearly the consolidation of this CounterTrend system into the composite mix of systems for the portfolio is therefore going to have a strong positive attribution to overall portfolio performance for this 6 year period.

So what does the equity curve and performance metrics of the 3 system portfolio look like?

I have included the consolidated result below.



Now we are starting to get somewhere. We are progressively beginning to flatten the equity curve which is having a dramatic effect on performance. Overall CAGR of the combined systems is 10.4% with a max drawdown of 20% (actually in two places) which is better than any individual system result. Furthermore we have had some great years such as 2014 and 2015 which delivered 33.9% and 45.35% per annum respectively.

Still there are weak spots in the equity curve that demonstrate that the selected range of systems is not sufficient to accommodate all market conditions during this period. Clearly 2010 and 2013 represent the weak spots for the portfolio.

So what you do now is closely examine the market conditions during these periods and seek to identify further system contributions that can target those weak spots while still remaining slight profitable over the entire timeframe. To do this you could elect to analyse different system methodologies or step into shorter term timeframes to examine the effect of the additional individual contributions to the overall portfolio.

While it is a fairly intensive data crunching process you can see how using sound logic and the approach recommended in the last few posts you can make your system more robust and capable of weathering the market storm while at the same time progressively improving your performance metrics without curve fitting.

What do I mean when I refer to the absence of curve fitting with this technique. The rules to this approach are fairly simple:
1. Each system must be sufficiently robust to be marginally profitable across the entire portfolio., In this case, the same systems are applied to all 46 liquid instruments. Position sizing is based on volatility based sizing methods where the same dollar trade risk is applied to each instrument.
2. All systems are turned on at all times and we do not attempt to preempt or predict market conditions; and
3. All systems are applied to 'almost' unseen data and all variables across any instrument or timeframe are consistently applied. A single pass is made across the data range to simply determine whether the logic of your system works and is sufficiently robust to produce a profitable result; and finally
4. Each contributing system must be equally risk weighted. It is tempting to cheat here and load a higher weighting to the better performing systems, but this defeats the purpose of not attempting to predict the market. Don't cheat...use equal weighting as I don't want to have to state 'I told you so' when future conditions do not replicate the past. You can never in principle predict what market conditions will be like tomorrow....... let alone next month.

While it is tempting to optimise your systems, it simply does not work when dealing with such an expansive portfolio, so don't even bother. What you need to do is design systems that respond to particular market conditions and have faith in the principle of low correlation to reduce your total portfolio risk exposure. We could get into the theory of this, but it get's complex suffice to say that this approach follows the Efficient Frontier Principle of Modern Portfolio Theory.

You can see that by applying this approach rigorously, you can continuously monitor and improve the performance of your systems and progressively make them more robust over time. Furthermore you can afford to take a break once the intense modelling is done and simply have faith in your fully automated system to take care of the rest....until of course rebalancing and portfolio walk forward time.

What this approach is revealing is that there is really not much alpha available in the market, so it puts a dampener on the ability to trade a single system for an extended period of time. Of course machine learning algorithms may ultimately be able to navigate this uncertain domain, but until then I have found that this approach is probably the most robust way to generating wealth creation in this market using automated systems.

Of course a skillful discretionary trader who has a range of techniques up his sleeve and can apply them in a similar way could in theory enjoy the same or better performance returns than the methodology described here...but.....and this is a big but.......how many skillful traders are out there that can apply a vast range of systems targeting different market conditions with the applied risk management conditions embedded in the framework. I would suspect we probably are only dealing with a handful. Anyway, that is a discussion for another day.

...that's about it........have a great trading week all. :-)

C