Diversified Trend Trading Approach

Hi Coper. Do you mind sharing the details of the systems that you used in testing?

Different people may have a different conception of what is long term vs short term and trend vs counter-trend.

Hi A

I work with a developer so can't give specifics without permissions but the general nature of each system is as follows:
1. Long Term Trend following system uses a Simple Moving Average on D1 of 200 periods with a Donchian filter to reduce overtrading and whip saws and simply exits on cross of the 200 SMA. An initial 3ATR stop is established to define R and the SMA becomes the trail thereafter. A very simple system simply used to capitalise on the long term secular trends and performs well during periods of crisis like 2008 and 2014/2015.
2. Short Term Trend Following System is designed to catch swings and simply enters after a set number of higher closes (long) or lower closes (short) with a 3ATR initial stop that then trails price and a take profit target of 2R.
3. The countertrend strategy is almost identical to 2) but after an extended set of higher closes or lower closes, simply enters in the reverse direction with a 3ATR initial stop and 3 ATR trail and a 2R take profit target.

They all operate off D1 so are probably longer term than most traders on FF appreciate but 2) and 3) have trade durations on average of 2-5 days or so whereas the long term trend following system can be in for months and even years.

PS We do not return profits back into trading capital in these examples and simply take any profits off the table. We could accelerate returns a bit further by fully reinvesting which is what you would do in a live fund management case....but not for the purposes of this exercise as we want to find the inherent performance of the system and risk profile on it's own terms before we look at ways to supercharge performance.

Thanks for sharing.

In my opinion, those rules don't sound very good and might not accurately represent trend-following. I'll leave it at that if this isn't going in the direction of the thread.

No that's fine mate. Both 2) and 3) are simply supplements to the main game which is the long term trend. The example above would not fit into the overall trend following space but I tend to be flexible in my options as opposed to rigidly steadfast like some of the CTA's I admire like Dunn Capital, Chesapeake etc....and your Mr Seykota etc.

What I haven't told you is that over a 20 year test system 1) takes the lions share of performance which dwarfs the other contributions from non-trend following systems. I use these supplements more for cash flow purposes as opposed to wealth management purposes. When running a fund *hopefully* it will be the cashflow supplements that will keep the investors off your back.

PS The last 6 years or so have been a fairly tough game for the Fund Managers (many of which are CTA's). Refer to performance stats below (not quite 6 years but you get the picture). You can see that this example is not too far off the pace when compared to the returns of the top FM's. Note also that their returns are net of fees so you have to take this into account with my example and discount the results accordingly. Mind you this entire example and the systems chosen were based on what was at hand for me this weekend. The reality is that a bit more science would go into the system selection than that described in the past few posts. It just turned out when I was crunching that the choice made performance progressively better.

I guess what im saying is not so much that they sound like ineffective systems but rather they don't match my definition of those terms.

To me, a short term trend system would just be a scaled down version of a long term system. In extreme cases, this also reduces the overall returns due to over-sensitivity to whipsaws. Having any kind of take profit point is already banking on reversion to mean.

Hehehehehe. Just defining a trend is problematic for me. Perhaps I should just term them diversified systems and leave the term trend out of it. But I get your point. The only system that let's profits run in this example is 1). The others respond to my greed in not letting some of those great unrealised gains just slip away. I will beg forgiveness from the trend followers out there for this pet aversion.:-)

Interesting analysis, thanks for the contribution Copernicus.

How does it fair with changing the SL from 3 ATR to say 2 ATR or 4 ATR ?

Not trying to find the best case scenario - hence no curve fitting. More curiosity regarding ATR SL size which will depend on market volatility.


The pic you have in post 512 i believe originated with Van Tharp. He uses bollinger bands to determine 'market type' and employs the associated strategy for that particular market type. I think Van Tharp has around 20 or so market types that he uses but don't quote me on that. There is another man - Sam Eder - who is a student of Van Tharp and has adopted the market type analysis as his bread n butter trading technique. Not a huge fan of his entry techniques but i don't mind his 2-10 and 2-25 position sizing (again originated from Van Tharp).

Morning Oz

You are an early bird. I assume you hail from Tassie.

With regards to the choice of ATR, it is personal preference which makes me stretch ATR to give a bit more breathing room for each system. A reduced ATR will leverage up my position size and potentially give me more bang for buck in performance..... but to shape it too close to normal noise doesn't assist me in determining whether the system has merit or whether it was just luck speaking.

Once I have a profitable composite of systems defined on their own merit with a reduced chance of being curve fitted, then I can test for small scale adjustments such as reducing or expanding ATR across the system and the entire portfolio and evaluate the material impact of the adjustment of the variable. If there is a performance improvement without a material impact on the metrics then I would be more likely to accept the adjustment.

The use of ATR itself in volatility position sizing is starting to become an invaluable tool for me. I am actually interested in Tom Basso's technique of using a EMA on the ATR to weight volatility towards the most recent price action. This way I can allow for the portfolio to quickly adjust for volatility and reduce overall risk exposure. Anyway, this is a work in progress for me.

Regards the pic....you are spot on. It was from Van Tharpe. Thanks for the heads up regarding his method of evaluating market conditions and for the pointer towards Sam Elder. Clearly from the examples posted above in this thread this method relies on the development of systems that respond to as many market conditions thrown at it as possible......so it will be worthwhile for me to crack open his books again regarding ways to evaluate market conditions.

C

Note on Interest Holding Costs - The Bane of a Retail Trader's Existence

It is worthwhile noting that the performance results of this example relate to leveraged forex instruments or CFD derivatives as opposed to say real futures commodities or even equities. The reason you need to consider stepping into the commodities markets or other markets if you get serious about actually making money in this game effectively boil down to transaction costs. Don't be misled by a focus on spread alone. The real killer to your overall performance is SWAP or other forms of interest holding cost.

To demonstrate using the example system performance ..........the total adjusted Gross Profit of the combined systems for the 6 year period (including spread) was $244,117 with a Net Profit (after holding costs) of $146,353 (CAGR of 9.6%). What!!!!!!.......this means that $97,764 (or a whopping 6.4% PA CAGR) of potential profit were the holding costs of trading these instruments. What a huge price to pay. No wonder it is so difficult finding an edge in this retail market. There is no level playing field in this game but we all know this...the brutal fact however is that the slope of this playing field is significant and is more akin to a casino's house edge.

That's why it is important to consider transaction costs and their impact on performance as you may indeed have an edge (with lower transaction costs) but just be unaware of it. An analysis of transaction costs should play a major role in influencing how you play this game.

If you want any chance of competing against the performance returns of the big boys, you really need to analyse your trading costs and think smart. Sure leverage is great....but there is a whopping cost to pay and there are alternatives to obtaining leverage (eg. futures contracts) where the costs are far lower.

Furthermore an addiction to MT4 as a platform has a nasty bite to it. No wonder the brokers love it (particularly the market makers) as it creates a dependency on products traded by MT4 that typically have significant costs embedded into the instruments made available to it by the brokers and the report outputs conveniently (probably deliberately) hide this impact. Full disclosure in reporting outputs would no doubt make retail traders less inclined to frivolously depart with their hard earned cash.....this applies to many other platforms as well and is not isolated to MT4.

C

Hey mate,

Im living in Canberra at the moment, never actually been to Tassie.

Thanks for the reply, not familiar with Tom Basso's technique using an EMA on the ATR. I assume he just weights more recent activity greater than the older data within the specified range (hence the EMA aspect).

Anyways, will be following your thread. I enjoy your content and the way you write errrr type.

All the best.


Cheers

Systems Diversification to Improve Performance - An Example - A Continuation

By now you will be getting the idea of how to shape your equity curve to respond to long range historical data to make your system portfolio robust and capable of handling the range of market conditions that have been thrown at you over the historical timeframe. If you have followed the rules, you will find that your individual system contributions to the portfolio are equally weighted and should all have relative low correlation with each other.

It is tempting to manipulate the weights of the system portfolio to produce as perfect an equity curve as possible but in essence by doing so, you are curve fitting to the market conditions and the timing of those market conditions by doing so. Where each individual strategy can alone be profitable over the back test, and provided each contribution is equally weighted, then your future distribution of returns and maximum drawdowns are more likely (statistically) to perform in accordance with the extended historical data.

The aim of the game is to plug the weakness in your equity curve over as long a market history as possible (that captures a broad range of possible conditions) by system additions that reduce the correlation of your portfolio as opposed to augment it.

You can see that this method is a game changer for those that want to have a role in managing their future risk-return profile and deliver long term sustainable results that are statistically valid as opposed to being mere 'hunches'. However many will simply not get it......and continue on in their search for an EA that delivers the golden egg. I wish them well, but make sure you turn the lights out when you are finished.

Obviously this approach requires a good understanding of a variety of systems and how they perform in different market conditions as it is this knowledge that will be the chief determinant of how you construct your desired portfolio equity curve.

Now rather than reinvent the wheel in your search for various systems to utilise, recognise that many bright minds have been involved in this game for hundreds of years....and all that hard work has already been done....in fact it is done almost every year by every new trader that gets attracted to this space......so here is a big tip in identifying suitable systems to code up for the purposes of applying these processes.

Get your hands on Perry Kaufman's treatise on trading systems and methods which includes specific details of hundreds of tested systems and describes the markets they perform in. Nearly every system you can think of is in there in all its glory.

Armed with this book, an EA developer or experience in coding, a database and spreadsheet experience and the principles described in this thread regarding system diversification....plus access to long term historical data, you should be able to design yourself some robust crackers that can stand the test of time. But remember it doesn't end there. At quarterly or 6 monthly intervals use the now extended time series data to do it all again and rebalance the portfolio. The most recent data may have conditions that have not been experienced over the last 20 or so years, so as time goes on you progressively increase the robustness of your system with new data. Don't use a defined timeframe for walkforward testing which rolls-forward. Use all the available data for each test.

While many think that testing over historical data is a useless exercise, for this endeavour at least, historical data is critical. It is your only guide for what may possibly be thrown at you in the future.

I will not let you get away with that, you are not forgiven, period.

Bass........Pleeeeeeassssse, let me out.....I wasn't thinking.

You know that i'm doing the same so all is forgiven.

Thank goodness. Now I can return to my chart. Now where was I? :-)

Hi C,

How do you think the DTT/EDTT can supplement (or complement) this diversified fully automated systems approach? I would think they belong to the system #1 category (and maybe system #2 as well, for the shorter timeframes). But then again, the DTT/EDTT is - as Robert Carver puts it in his book - a semi automated approach and not fully automated. So is there a way to mesh it with a counter trend fully automated or semi automated strategy? Or perhaps would it be better for the DTT/EDTT to be used as a standalone trading system? Just want to hear your thoughts on this

What I found interesting in your posts is that “systems should run at all time”. If I got it correctly, what you’re saying is that a trend-following system should run no matter what the market condition is, and at the same time, a counter-trend system should likewise run no matter if market is trending or in a sideways movement. Put these two to run together and this should smooth the equity curve, while maintaining a positive non-random bias. In the past, I had always thought — and in fact, find it difficult — that identifying a market condition as trending or in a sideways movement is essential to know what type of trading rule I will fire off (trend or counter trend). Hence, this is interesting…will have this as food for thought in the next few days to see how viable this is for me.

Cheers mate!

Hi G

I think you are spot on in your views. The DTT/EDTT clearly performs superbly in 4 of the 6 market conditions defined by Van Tharpe yet clearly it's weakness lies in volatile and non-volatile sideways trending markets. However the stringent risk management approach adopted by this discretionary system and the effective placement of predictive entries away from normal volatility acts like a Donchian filter and prevents over-trading during these sideways markets...however as we can see in the results this year for example, it's not perfect and symptoms of drawdown are appearing particularly in the EDTT approach.

Unfortunately as we can't get this fully automated it is very difficult to conclude how this would perform over extended periods of 'no trending' conditions but I suspect that a 20% drawdown during say 2010 to 2013 may have been possible.

Just as a safety measure and given that these methods perform superbly during trending periods, we could afford to slightly lower our expectations of overall profitability of the DTT or EDTT using current trade risk %'s, and play on the safe side by including a mean reverting system into the total equation. Provided we can manage the potential drawdown of the total system we might find that we can achieve a better long term result with lower risk allowing us to scale up our trade risk % across all component systems without adverse portfolio risk impact.

I don't think it would be a problem running a fully automated mean reverting system in conjunction with this semi-automated approach, and in fact it certainly would assist in not unnecessarily complicating the trade management process further. As you will be aware things can get quite busy during trending times given the number of timeframes we are operating on. It is simply a matter of developing a dashboard to manage and monitor the semi-automated and automated processes to make life easier.

It is just a darn shame that we can't fully automate the entire process....then we could let it trade while we sip martinis with limited intervention.

I hope you are enjoying the Carver book mate. It took me a while to get into it but after pursuing with it, it was certainly a worthwhile effort.

Hope all is well.

C

Monthly Performance Results - DTT/ EDTT/ EDTT1 - November 2015

Below are monthly performance results for the various discretionary trend trading strategies which are the subject of this thread and the regular music supplement (i just can't help myself....you all know that)













Commentary
Despite 2015 being a bleak year for trend trading in general, it has been a valuable exercise as it allows you to examine the performance of these strategies under a bit of stress. Clearly the glorious result for 2013 and 2014 were abnormal for trend trading as I suspected, so at least a bit of reality is now entering the game where we can start putting the risk management aspects of the strategies to the test.

Year to Date (YTD), the DTT is pulling the best results, not in terms of profit but in terms of Profit/Max Drawdown which is to be expected as it's trade frequency is far lower than the alternatives and the year has been a shocker for the trends. The DTT has managed to deliver a 12.2% return on capital over the 11 months to November with a max drawdown of 3.13%. The EDTT1 (no Countertrend trades) so far for the year is in second place with a return of 13.3% and max drawdown of 7.65%. In last place is the old nag (EDTT) with a 14.1% ROI and drawdown of 9.54%.

You can clearly see that a protracted sideways market of say 1-2 years extent would put the EDTT and EDTT1 strategies into more severe drawdowns , which suggests (based on recent discussions in this thread) that you may need a reversionary strategy working in tandem with these strategies to provide a degree of downside protection.

Despite the solid result of the DTT, it is hard to not get excited about the performance potential of the EDTT and EDTT1 when the markets start trending....so they definitely generate significant profits provided that the trends are not too few and far between .......which can be used to subsidise some enhanced risk performance measures to defend against protracted non-trending periods.

Note that the results detailed in the tables below represent closed trades. What must be noted is that favourable trending conditions commenced in earnest in November resulting in considerable unrealised profit in the portfolio that will be beneficial for December. With a bit of hindsight talking based on the fact that I am posting now (on 15 Dec), a lot of that unrealised profit from the shorter term timeframes vapourised with the twitter debacle of the Financial Times in early December......however despite the pain of seeing that stockpile vapourise, some excellent trends are still alive and some of the indices and CAD pairs etc are firing on all cylinders. Hopefully December will finish the year off with a cracker month. Time will tell.

Current unrealised profit in the portfolio based on current valuations is detailed below. There is almost 38 R of unrealised profit sitting in the portfolio at the moment with great contributions coming from CADJPY, EURCAD, FTSE, NIKKEI, NZDCAD and USDCAD. God bless you Canada. I always loved your Maple Syrup. Let's hope that these unrealised gains are not blasted out of the park as experienced in early December with that fat fingered fool of the Financial Times.



PS The performance of the participants of the Allegris index is starting to fire up in November with a cracker result and knowing the capabilities of some of these fund managers, they will be roaring home in December so watch that space.

Intelligent EA's or Multiple Systems?

As most can guess the flavour of this thread is ultimately converging on fully automated systematic trading and we have recently explored a method of applying multiple systems as a basis for portfolio diversification to allow a composite of EA's to most comprehensively address market conditions over an extended history.

A question therefore comes to mind in whether this "pre-fit" composite system approach is a more or less preferable exercise than marching along alternate machine learning methods of problem solving.

I thought I would have a look at the pro's and cons of both possible routes towards more effective EA development.

Machine learning algorithms effectively employ a walk forward statistical processing technique where patterns associated with most recent market conditions are progressively assessed in terms of their repetitive frequency (pattern recognition) and thereby allowing the EA to iteratively adjust it's input variables to respond to the most prevalent features of current market conditions. This process is thereby a moving feast of progressive adjustments that take time to manifest as responsive trading variables....hence there is always a significant lag in responding to current market conditions. There are however movements afoot in the machine learning world to hold past analytical results and responses in memory which can be quickly brought to bear if conditions arise that suit a previously solution that was successfully applied to a particular past market condition. A further benefit of this approach is that it is quicker to respond to new market conditions previously unobserved over history that other alternate methods.

A multiple systems approach to portfolio composition however has already embedded in the portfolio a range of systems that have demonstrated their capability to respond to past market conditions, hence the market memory is already embedded in their system design. Obviously a problem and weakness of this methodology arises if new market conditions emanate that haven't been observed in the past, and thereby rendering the system composition of the portfolio obsolete and too lagging for navigating new waters. It is hoped *crossing fingers* that utilising a long historical dataset that encompasses a wide variety of past market conditions is sufficient to stand the test of time....but there still is the assumption that 'the past is a guide to the future' embedded in this logic.

There is a third approach to EA development that utilises a reverse engineering approach to it's development. Simply plug in the performance metrics desired for a system and then let the EA create it's own code to achieve this result. An interesting podcast surrounding this principle can be found here. Obviously the question therefore is, how do you avoid curve fitting with this solution, and there are a number of strategies that can be employed to address this issue. EDIT: There is also a nice link about trading the equity curve here that demonstrates some techniques that can be employed with algorithms.

Despite the pro's and cons of each method the strategies developed to overcome the problems associated with algorithmic trading are quite exciting as you can see how the processes involved are steering away from linear programming styles towards parrallel processing or multivariate responsive programming techniques that are taking their queues directly from evolutionary adaptive sciences and the cognitive neurosciences which far better exploit the domain of the discretionary trader and that of the human brain.

The direction being taken certainly points to the fact that the skills of a talented discretionary trader will shortly be inundated by the machines. We have seen the exchange trading floors change form with technologies and the exponential growth of automated trading with HFT. I therefore assume that humans are shortly going to have to move out of this space and alter their professional aspirations towards system monitoring and management as opposed to the rapidly dying art of manual trading. There are no certainties in this world apart from one particular enduring feature of this universe..... 'that nothing is constant and everything changes' . In that sense it pays to be flexible and ensure you do not become wedded to a particular technique or style as it is unlikely that it will be an enduring feature of this market. Will trends persists??????.....I think we need to keep an open mind here despite the 800 or so years of history of its prevalence. I err on the side of yes...but recognise that an iceberg can sink an unsinkable ship.


Now listen kids. Don't touch the displays. This was what a trading floor was like in the olden days.

"Nowhere does history indulge in repetitions so often or so uniformly as in Wall Street. When you read contemporary accounts of booms or panics the one thing that strikes you most forcibly is how little either stock speculation or stock speculators today differ from yesterday. The game does not change and neither does human nature."

- Jesse Livermore