Systematic Trend Trading

Which parameters would you randomize in your Monte Carlo simulation? Do you test for robustness in market / execution problems only (slippage, spread etc.) or also vary you indicator parameters? Which do you think are the most important parts to randomize?

Hi C

I trade on the D1 timeframe or above so the impact of slippage and spread variation is minimal. I can see that in Monte Carlo runs. Obviously spread and slippage become concerns as you trade the lower timeframes. That is a good reason why to focus on higher timeframes.

Spread Variation: The charts below actually is representative of 200 runs with a 20 point variation.



Slippage:



With trend following strategies the biggest variation in array is by either randomising strategy parameters, Randomise skip position entry or by randomising the backtest starting bar. The latter two variations can be easily understood in trend following world because you never know when you are going to catch the biggie. Results are not evenly spread throughout the data set....so if you miss a small number of them or start your trading at a different time and miss a few of the outliers, then the result on your equity curve can be significant.

Position Entry:

Attached Image

Backtest Starting Bar:

Attached Image

The key variables in regards to randomising strategy parameters for a trend following strategy that have the most influence on the array are:
1. Initial stop variation; and
2. Trailing stop variation.

The variation of entry parameters have far less variation on overall results. We have already discussed this on this thread where we acknowledge that even a random entry can work. It does help however to have filters to remove your strategy from market noise and also choose the direction of the primary trend.

The key ingredients are the initial stop for both lying outside normal ATR which allows you to cut losses short, give room for random noise to operate and also to provide for volatility based position sizing purposes and the trailing stop which allows you to let profits run.

To be honest C....after I have told you all this....I now actually do not use Monte Carlo, walk forward segmentation or any other data mining technique as they are better suited to prediction methods rather than price following methods. The MC array is however useful to illustrate points of interest and identify where the secret sauce lies in the key parameters to look at.

The best method to test the voracity of your system is data and the more the merrier. Forget Monte Carlo and Walk Forward in this trend following space. They are not worth the effort. The best way to assess whether your strategy is curve fit or not is to test it on say 10 years of unseen data. FULL STOP. These other work around methods are not terribly efficient and do not equate with how a price follower such as us see how this market works. Trend following is not a regular occurrence where you can expect regular cashflow. It is a wealth building exercise based on outlier events that cannot be predicted in advance. As a result, we are after a stepped equity curve profile and not a pretty linear one.

The best advice I could give for us price followers (not predictors) is to use at least 20 years data. Develop your strategy and backtest it on the first 10 years....then simply test it unseen on the next 10 years of data.

The aim of price following is to catch market transitions. You do not need finesse in your entry to achieve this. Transitions do not require a solution that is built with high precision. You need very simple strategies that have lot's of room to breathe to allow for volatility during transition but observe the key points of initial stop and trailing stop condition. There is more than enough fat in outliers to avoid preciseness which is rarely repeated. Too tight and you will introduce unnecessary drawdown into your equity curve. You may get whoppers but otherwise death by a thousand cuts will kill you. There is a tradeoff here as market transitions are unpredictable and you can be waiting a long time between drinks.

In your 10 year blind walk forwards test you will soon see what stacks up and what doesn't. You will also realise how slim an edge we actually have to work with and why diversification and the 'outlier' is so essential to the cause. Above all else....don't be fooled by randomness.

Cheers

Example of a Backtest and Walk forward Process

Below is a backtest on EURUSD from 1 Jan 2000 to 31 Dec 2010 using a filtered Donchian Breakout System. A lot of work is undertaken during the backtest process to filter what is regarded as non-trending zones. To do this you superimpose market data over the backtest.

Like this



Notice the following in the equity curve above. I catch some good trends (represented by the balance line jumping up)....but also miss a few of them where the balance line is flat while the market is trending. The reason for the misses are the filters applied to keep me out of the rough (swings and roundabouts guys....you can't avoid it).

But importantly notice how the strategy performs during less trending regimes. You will see that losses are mitigated or that the strategy simply does not trade during these periods. Also note the max draw over the IIS period of 0.75% (remember this number for later).

Now that I have configured my strategy according to this backtest to risk mitigate and also catch some good outliers I can unleash it fairly confidently into the unseen world without worrying about it falling off the cliff. While I don't know what the future holds I do know this is configured to catch occasional outliers but more importantly....will protect my capital during adverse conditions.

Let's see then how it rocks on unseen data up to today.



As a trend follower I am very happy with the OOS result. Notice how it snagged the 2014 bear trend in a big way. Also note how capital was protected during fairly low vol and sideways movement between 2015 to current day. The strategy is therefore sufficiently robust for me to take forward as a good inclusion in the trend following portfolio.

Now refer to the Max Draw in Jan 2012 of 1.81%. Double the previous max draw. This is just the nature of the game guys. Many would have turned off the strategy with a double sized drawdown...but then you would possibly have missed the huge outlier of 2015. You have just got to accept that your worst max draw if always ahead of you as the data sample gets bigger. Furthermore you have got to accept that being in drawdown is just a fact of life for a price follower.

What is most important however is to examine how long you can stay afloat during adverse market conditions. This is why a filter is so important for price followers to avoid the market noise and stay out while markets are not trending.,,,,and this is why backtesting is so important in this game. Without it you will be steering blind into uncertainty.

great posts C.... appreciate your time and effort.

Interesting comparing our two approaches, you filter out a lot of conditions and I have been aggressively trading through all conditions. (although I do run a more filtered strategy to diversify)I take on a lot more volatility and have to deal with that by reducing position size. My gut feeling is that mixing high and low volatility strategies is a good plan, although if I had to assign a number to that I would say im 65% certain, I wonder what your thoughts are on that?

Hi 60. Thanks mate. I hope it doesn't detract from your great thread. If it does then tell me to pull my head in. I am tending to inhabit your thread more than mine...but it is great to have like minds (trading philosophies) collaborating a bit.

Yep..I am thinking along your lines as well. If we can merge both types to add further diversification and up the trade frequency a bit we will literally clean up :-)

I still have got to get my head around being market selective. My approach is market neutral....but your approach definitely has legs. How discretionary is your system 60 and have you coded it up at all? I have a feeling that there is a bit going on in your decision to pull the trigger that is really helping you. Plus your addition with new signals after each week elapses is probably the secret sauce in your technique that allows you to significantly take advantage of big trends when they happen and pay for your whipsaws from your increased trade frequency.

Let's assume that it is the outlier where our wealth is made. I catch fewer outliers (due to filters) than you with single positions but also have less whipsaws to worry about. You catch all outliers (going for all of them) and add to the trades along the way when they are of sufficient duration but need to pay for more whipsaws. My Win% is around 40% while yours is probably more like 20% or so.....and yet both methods seem to have a strong edge. That gives me hope that there is a blended solution in there at the portfolio level which will significantly improve our risk weighted returns over the long haul.

not detracting at all C, love your posts on here. I worked for 15 years as a record producer and then retrained as a chef, I come from an art background so my approach is a bit more fuzzy, I'm assuming you come from a more scientific background so your posts add some clarity to the ideas. Same theory from different approaches.

This strategy started out 100% systematic and was coded so I could run enough data (luckily, I know some great programmers who went above and beyond crunching huge amounts of data for me). I was never really looking to optimise but to discover what factors really matter. Does entry matter, does filtering conditions matter, exit, stop loss, instrument etc etc. What I was looking for was confirmation that simple systematic trading works and more importantly how to make it work. Can you distil everything down to a single statement, a single idea to base all of your trading off?

This is what I was left with:
As long as you have asymmetry in your risk-reward and position size allows survival over very long runs, you will always make a profit.

Those are the important elements and the rest is trade-offs between higher or lower volatility. I guess we all want more precision than that, more certainty, but I'm not sure I want to spend the rest of my life playing that game (not even convinced it's a winnable game) As we have discussed here, you are better of picking a mixed bag of settings than looking for the 1 anyway

So this system is not discretionary, but the lack of precision in certain areas makes it appear so. Last week I made the mistake of trading short when the signal was long, doesn't matter, Sometimes I will stay up late and trade the Asian session, or if I am staying in Asia, doesn't matter, sometimes I will place my stop at the high, sometimes at the top of the body, doesn't matter - you can't control the variance so why bother. Traders made money before precision because they had simple big ideas they followed with good risk control and I am looking to emulate them.

*Don't be afraid to shoot any of these ideas down, I'm always open to opposing opinions in life

Spot on 60

I think it is as simple as that. As much as I like to get sciencey with this stuff....it all boils down to a simple feature of the markets. They have 'fat tails'. They cannot be faithfully described by a normal distribution. They are not totally efficient. When coordinated behaviour rises from unpredictable 'events'.....the fat tails rise and provided you have an asymmetrically designed EA that cuts losses short at all times and is unbounded in profit target....then you are virtually guaranteed to win in the long haul provided you are patient and can suck in the drawdown. Diversification helps to increase the frequency of fat tailed benefits and reduce the long wait for these events and also importantly suppresses the drawdown from uncorrelated return stream additions. Everything else we do around the edges is 'swings and roundabouts' :-)

Cheers mate, Great to be here :-)

Talking about outlier trades, this one on October 3rd could still be going strong! Any thoughts?

Yep, it's surprising how many trades do go to unthinkable returns, you have to decide where to lay your hat and be ok with that trade off. I remember Tom Basso commenting on where is the best take profit and his answer being "never". If I was going to run a system with really large RR I think I would go to equities though - as forex is trading one instrument against another it tends to have smaller ranges in general. I have run variations of this system with a 10R trailing stop though and it was actually very good - but it did highlight to me that running a 10R and a 5R trail at the same time was way more robust.

A great way to deal with the trade-off is to diversify and trade a range, as a crude example: one system that trails stops around 2R, one around 5R, one around 10R, or even splitting positions. Sounds too simplistic but if your aim is to compound a small account to something large it can give the consistency over time needed.

Welcome to the thread and thank's for posting

All systems performing well at the moment and hitting new equity highs, in the past I would attribute that to 'good trading'. The reality is the markets are just giving me the conditions my system is looking for. I actually have no control but that can be a comforting feeling. Let go of the desire to control profits and just deal with risk.

I recommended this book earlier in the week, really enjoyed it, why we all fall for the con even though we know we're being conned

https://www.amazon.co.uk/Confidence-...gateway&sr=8-1

Don't understand...could you elaborate even further...

Let's say you run tests on a system with a trail at 1R, 3R, 5R and 7R. You look at the results and you really like 7R because it has by far the highest returns over the test. At the same time it has the highest drawdown and most consecutive losses. 1R and 3R have really low drawdowns but the total returns are modest. The high R system is clearly more volatile and from a risk perspective you need to reduce position size, the low R versions are much less volatile and you could therefore increase position size on them.... so really what is the difference, how can you pick the best? Simply put, the best option over the long run would be to split your funds to the high and low volatile systems and not select just one - don't put all your eggs in one basket, if you like.

This works on the numbers and psychologically...

more consecutive losses and more drawdown means more returns, trading is weird like that

Indeed, and really important to understand

Precision is curve fitting. Make your rules loose, trade weak signals, look for systems that generate lots of small losers and small winners, these are the systems most likely to catch the big winners.

The more precision you add the less likely your system will be able to adapt to future conditions

The old risk to return chestnut. Higher returns with higher risk where risk is expressed as potential value variation per unit of time (aka as a measure of the volatility expressed in your total equity allocated to a system). Risk is not the distance between our entry and our stop. This is simply potential maximum loss exposure. You could have a wide stop and wait for 2 years to reach this level....or you could wait for 2 minutes. Clearly risk is different in each of these cases.

By viewing risk this way you can understand why an allocation of $10,000 to Bitcoin is of higher risk than an allocation of $10,000 to US Treasury Notes.

Risk should be viewed as the degree of volatility in an assets price or the degree of volatility in your equity allocated to a strategy per unit time. This catches many of us up and why you need to look at risk-adjusted metrics to select from different risk:return options.

For Trend Followers a great metric to use is the MAR ratio of something similar. Sharp is useless as a measure for trend followers due to the natural volatility of our equity curves and is better applied to assessing mean reverting systems with nice linear rising equity curves while they work.

MAR = CAGR %/Max Draw % = Annualised compounded return/Maxmimum Draw%

or you can also use

Total Return%/Max Draw %

By expressing the relationship between return and drawdown as a ratio, you then select the highest ratio as the preferred risk-weighted solution. It helps to standardise the risk to reward relationship across different systems and allow for a direct comparison between options.

Once you choose the best Risk adjusted solution then you can scale it up or down using position sizing to adjust leverage of the total outcome.

For example using 60's stats above for the 200MA option you can scale your position sizing and adjust it to reflect a desired drawdown and also select the most suitable option from a risk adjusted basis.



If it was me....I would be looking at EUR9R option as a preferred risk adjusted system for the 200MA.....and then I would increase my diversification wider to other markets with the 200MA system to make up for not having the EUR3R and EUR5R and EUR7R options (reduced trade frequency) by choosing uncorrelated systems that offer similar high risk adjusted returns (that lift trade frequency but of an uncorrelated nature).

What you will find is that all the EUR 200MA options are fairly highly correlated. By trading all 4 of them, you are effectively multiplying your position size by a factor of 4 times across a correlated suite of returns. Also the variation between 100MA and 200MA will make the EUR between these two versions fairly highly correlated.
This will exacerbate drawdowns when conditions are unfavourable.

So when having a range of correlated systems to choose from, select the single option that offers the highest MAR. This will then force you to look at other uncorrelated systems to increase your desired trade frequency. You can then adjust all strategies using this risk adjusted measure to achieve a desired drawdown per system and risk weight position sizes accordingly with the knowledge that the uncorrelated nature of each system will reduce your overall drawdown at the portfolio level.

What we are discussing here is what you do at the portfolio level to achieve best bang for buck in all your systems, ensure a steady growth in the equity curve at the portfolio level attributed to the uncorrelated nature of component systems (reduce the wait between drinks) and also manage the drawdowns at the portfolio level through uncorrelated addition of systems.

If 60 had 16 uncorrelated systems all appropriately risk adjusted to achieve a desired 20% max draw....then the overall portfolio will have a max draw of far less than 20% and strong returns spread more evenly at the portfolio level. This would be a preferable alternative to having say 4 x 4 systems in which each grouping is fairly highly correlated and also across the 100MA and 200MA options (which will also be fairly highly correlated).

Having wide diversification (systems, timeframes and markets) helps to reduce the psychological impact of many consecutive losses in a single return stream so you can focus on lifting your performance. Usually something in your diversified portfolio is reaching for the stars to take the mind off the many whipsaws.

Note that the Estimated Gain% is an approximate only particularly if you compound the account. The nature of consecutive wins and losses and when they occur in the time series can influence this estimated result but as an average it is a pretty good proxy.

I use my short term system eg. 4R,5R,6R exits just to smooth equity curve or keep my head above water when markets are choppy. I make real gains holding for 400R,500R,600R exits.

Wow. What monsters :-) You are a patient trader. Do you trade forex and deploy a trail?

I just use a monthly chart to move stops underneath higher lows in uptrend and lower highs in downtrend. That's it, nothing fancy. I have a few short EUR/USD positions in large paper profit , bracing myself to give a fair chunk of that back when I eventually turns north.

Thanks D. Interesting how profitable it gets as you go further out in timeframe. Reminds me of Nick Radge's weekend trader that makes smaller timeframe trading seem a waste of time. It keeps the brokers happy though the more you churn :-)