Diversified Trend Trading Approach | Page 183 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Is Grid Trading (combined trend following and counter trend) Profitable? 3 replies

Making a good diversified portfolio 10 replies

Diversified EA Money Management Code 10 replies

HAMA PAD - A Simple Trading Approach 7 replies

Trading a (semi) automatic approach...but want more 11 replies

Attachments: Diversified Trend Trading Approach
Exit Attachments

Diversified Trend Trading Approach

  • #3,641
  • May 17, 2018 7:36am May 17, 2018 7:36am
Quoting PipMeUp
{quote} In a trending market the price gets far from a MA. If the lag of your MA is constant at all frequencies (linear phase) and if the trend is constant (straight line) you can show that the distance is proportional to the trend. Even if it is not the case, you can estimate the mean distance of the price to the MA and its variance (in a rolling window). Because the MA lags below/above the price by the average value you can adjust by adding the mean to the MA. Surround the MA with 2 and 3 SD. The slope of the (original) MA over SD (slope/sd) gives...
Many thanks for this info Pip. It really helps.

Mate...no sign of your email. Do you want me to PM you and give you my email addy again? In the subject heading type PipMeUp and I will check my spam if I still get nothing. Sorry about this Pip :-)
  • #3,642
  • May 17, 2018 7:41am May 17, 2018 7:41am
Quoting thomasK
Since we throw KAMA ideas on the table, my previous post concerning NVDA gave me the following idea. As we can see in NVDA, during those last 4,5 years price never even touched KAMA 10,5,30 So apart from looking for KAMA crossovers, how about using just price crossovers with KAMA 10,5,30 ON TWO TIMEFRAMES (otherwise we should go back to elementary school ). Entry according to the direction of the higher TF, entries and exits on lower TF. Here is a relevant screenshot of DOW. tk PS The area in the rectangle...
stock markets have an upward drift, in the last 8 years or so central banks printed money and took long positions in stock markets (like snb with apple)
using patterns/tricks that work in forex should work also in stocks (in fact some of them better) but going for exemple short in stocks using same signal/trick for going long, using same rules for placing stops (distance from your entry to your stop point) is not a good ideea
even trading breakoust, one of the easiest patterns and robust should have different setups/rules for bearish markets
using intra-day charts for trading stocks will lead you to the poor house. (just think about risk management, in us they are like more than 10.000 stocks, defining some rules to pick from that list maybe will let you with 400-500 stocks to trade in every day, even like that they are to manny, what about gaps, or if sec will ban one company?)
using intra-day charts for trading stock indicies is a different story (just remember again that stock bearish markets are different in instinct/accting than bullish one)

using heinkin ashi or how that charts are called will not help you (just my opinion)
study normal candle/bar charts again and again, you will find some patterns/tricks that can generate 20x or more your initial risk even on daily charts not at the top/bottom of the move but even in the middle of it after you defined a trend
Price itself it's an indicator
  • #3,643
  • May 17, 2018 8:23am May 17, 2018 8:23am
Sorry to disturb your conversation, I did not read this thread only the title. Did any of you guys try an approach of some kind of MA and make the period based of a function of price movement. Define the period of the MA by N, then:

e.g. when N = alfa * sum(P(t)-P(t-1))

Where the sum is a geometric sum for example. And alfa some shaping parameter. So the main goal of this is when price is not trending we get a big N resulting in that the "signal" is "flat" when not trending, when there is a trend it will closely follow the price since N will be small.

Probably this will show a nice price following curve, but not having really any predicting power.
  • #3,644
  • May 17, 2018 8:32am May 17, 2018 8:32am
Inserted Video
Cut short your losses. Let your profits run on. David Ricardo (1772-1823)
  • #3,645
  • May 17, 2018 9:00am May 17, 2018 9:00am
Quoting Jysticim
Sorry to disturb your conversation, I did not read this thread only the title. Did any of you guys try an approach of some kind of MA and make the period based of a function of price movement. Define the period of the MA by N, then: e.g. when N = alfa * sum(P(t)-P(t-1)) Where the sum is a geometric sum for example. And alfa some shaping parameter. So the main goal of this is when price is not trending we get a big N resulting in that the "signal" is "flat" when not trending, when there is a trend it will closely follow the price since N will be...
This is more or less what KAMA does.

For the MA in this post https://www.forexfactory.com/showthr...5#post11069935 I do this:

The Goertzel algorithm measures the power at a specific frequency in a discrete signal. I build 74 such filters each tuned to a period between 2 and 75. At the same time I compute all the SMA of period 2 to 75. I compute a weighted average of them according to the power measured for their period. I get a self adjusting MA. I do the stats explained in the post and post filter with a EMA which alpha varies between 0 and 0.8 according to the slope of the GoertzelMA divided by the STDDEV. alpha = min(0.8, abs(ma_slope / StdDev))


This is the GoertzelMA without the post EMA filtering:
Attached Image (click to enlarge)
Click to Enlarge Name: GBPUSD-H4.png Size: 25 KB
No greed. No fear. Just maths.
  • #3,646
  • May 17, 2018 3:34pm May 17, 2018 3:34pm
Cool, still not really predicting power, however you shouldn't expect that from such things right? Probably don't understand your concept fully yet, however could be interesting weighting oposite price movement more.
  • #3,647
  • May 17, 2018 4:19pm May 17, 2018 4:19pm
Quoting m3rlin
{quote} stock markets have an upward drift, in the last 8 years or so central banks printed money and took long positions in stock markets (like snb with apple) using patterns/tricks that work in forex should work also in stocks (in fact some of them better) but going for exemple short in stocks using same signal/trick for going long, using same rules for placing stops (distance from your entry to your stop point) is not a good ideea even trading breakoust, one of the easiest patterns and robust should have different setups/rules for bearish markets...
Nice M. Totally agree mate and it appears that the heavy duty Fund Managers in the diversified trend following space also agree with this verdict regarding equities such as Nick Radge, and Andreas Clenow. They simply refuse to adopt short strategies in equities and most tend to adopt rotational absolute momentum strategies in this space. Clenow's book "Stocks on the Move" gives some great examples on how to deploy this natural long only bias in equities.

Equities is certainly an asset class offering some great opportunities and there is more and more interest being expressed by the CTA's in this segment as part of a diversified portfolio on their incessant search for diversified opportunities.
  • #3,648
  • Edited 5:06pm May 17, 2018 4:30pm | Edited 5:06pm
Quoting Jysticim
Sorry to disturb your conversation, I did not read this thread only the title. Did any of you guys try an approach of some kind of MA and make the period based of a function of price movement. Define the period of the MA by N, then: e.g. when N = alfa * sum(P(t)-P(t-1)) Where the sum is a geometric sum for example. And alfa some shaping parameter. So the main goal of this is when price is not trending we get a big N resulting in that the "signal" is "flat" when not trending, when there is a trend it will closely follow the price since N will be...

Quoting PipMeUp
{quote} This is more or less what KAMA does. For the MA in this post https://www.forexfactory.com/showthr...5#post11069935 I do this:
Hehehe. I love it when you get a merging of the minds in this forum. It reveals the central nut of the problem that many of us are trying to ferret out....."how to discriminate between signal and noise and then how to apply it in practice".

I love your approach to this Pip. Heavy duty stuff. The fidelity of that indicator is impressive mate....and probably a more effective approach than Perry Kauffmann's. I was clearly seeing your point regarding your issue with KAMA simply going flat in congestion....however it did save my ass from quite a few whipsaws by doing this in a crossover arrangement and that made this indicator appealing to me. Have you developed a crossover technique with your Goertzel indicator? I am wondering whether this indicator solves that issue with Perry's?

Your posts have focused my attention on how to avoid the impact of outliers in a dataset and perhaps a 'threshold' discrete as opposed to continuous approach to indicator development might be an apt way to deal with this. It certainly tends to work with 'non-linear problems' such as neurons and transistors in filtering out the noise.

In other words an indicator that remains flat until a certain threshold in a weighted average vertical price movement lifts it over the threshold. You could probably achieve this to a degree by choosing a method that applies a threshold indicator which operates over a variable time frame defined by the 'signal to noise' nature of it's underlying dataset *hoping this makes sense*. That way you have an adaptive mechanism to address non-stationery market conditions. I understand this is what Perry was trying to achieve in his indicator development

Just thoughts at the moment trying to reveal the central problem we are trying to overcome. An issue I am currently facing whether by SDC or other approaches and what interested me in KAMA is attempting to choose an appropriate length of segment to apply for an assessment of the noise to signal.....and this attempt to define a universal discrete dataset interval is probably the problem I need to pull apart for non-linear conditions. Maybe the data should be defining the appropriate length for me and while being discrete can extend and contract defined by market conditions.
  • #3,649
  • May 17, 2018 4:34pm May 17, 2018 4:34pm
Quoting Ramadas
https://www.youtube.com/watch?v=hb-nwXhglpM
This makes you want a Ford Mustang. Time to visit Harlem :-)

Inserted Video
  • #3,650
  • May 17, 2018 5:16pm May 17, 2018 5:16pm
While I was Sleeping

Attached Image (click to enlarge)
Click to Enlarge Name: Capture.JPG Size: 118 KB


The markets are awaking from the doldrums. A bit more activity in the portfolio. How is your dynamic universe going R? The red lights and alerts on your console would be keeping you active.

Attached Image
  • #3,651
  • May 18, 2018 1:22am May 18, 2018 1:22am
Quoting Copernicus
While I was Sleeping {image} The markets are awaking from the doldrums. A bit more activity in the portfolio. How is your dynamic universe going R? The red lights and alerts on your console would be keeping you active. {image}
Yep, always something going on:

Attached Image (click to enlarge)
Click to Enlarge Name: JGO4DXo.png Size: 38 KB


Attached Image (click to enlarge)
Click to Enlarge Name: vvpudcK.png Size: 63 KB


Attached Image (click to enlarge)
Click to Enlarge Name: IPNoQFV.png Size: 20 KB
Cut short your losses. Let your profits run on. David Ricardo (1772-1823)
  • #3,652
  • May 18, 2018 2:51am May 18, 2018 2:51am
Quoting Jysticim
Cool, still not really predicting power, however you shouldn't expect that from such things right? Probably don't understand your concept fully yet, however could be interesting weighting oposite price movement more.
When you filter a sin wave with a SMA of exactly the same period the positive values of the sin cancel out with the negative ones. The output is flat. If the signal is a trend plus a sin wave you recover the trend. Of course if there are more than one sin waves you should filter twice to recover the trend. Unfortunately the lags would sum up... Goertzel is a way to estimate the strongest sin wave to filter out. But Goertzel also lags a bit and there happens to be two cycles in force at the same time. The averaging of all the SMA weighted by the strength of the response of the Goertzel filter bank is a pragmatic way to mitigate the problem.

The greenish indi in the picture is the periodigram of the market at each bar. The lighter the green the higher the energy at this period (from 2 below to 80 at the top). The red line tracks the highest energy. You can see jumps where there are two big cycles.
Attached Image (click to enlarge)
Click to Enlarge Name: periodigram.png Size: 82 KB


Quoting Copernicus
Have you developed a crossover technique with your Goertzel indicator?
Originally my idea was to detect and bracket the flat markets to play convergent strategies. The boundaries of the envelope give a definitive stop beyond which you should consider a breakout is probably happening. I never succeeded in building such a strategy. The always end up to hold two of three positions in lose at the BO time which are enough to wipe the few pips collected in the range. => ranges are too short lived.
No greed. No fear. Just maths.
  • #3,653
  • May 18, 2018 3:12am May 18, 2018 3:12am
Quoting Copernicus
An issue I am currently facing whether by SDC or other approaches and what interested me in KAMA is attempting to choose an appropriate length of segment to apply for an assessment of the noise to signal. [...] Maybe the data should be defining the appropriate length for me and while being discrete can extend and contract defined by market conditions.
I love it when there is no parameters in an indi. When everything is data driven.
- Mann-Kendall Test tells you the probability that some monotonic trend, non necessarily linear, exists in a dataset.
- Theil-Sen estimator tells you what is the slope of a linear trend if there is one.
What about scanning the recent dataset with all windows of size say 5 to 30 in search of the best fitting linear regression channel using Mann-Kendall and Theil-Sen? The top/bottom 2 sigma give you a way to trail a stop-and-reverse. If the noise is not fat tail you can bracket the swings and build an automatically tuned pSAR. Too bad it does work out of the box... I had to add many heuristics . Also it only works in range bar charts because time charts have fat tail distributed returns.
20 pip range bar chart of E/U
Attached Image (click to enlarge)
Click to Enlarge Name: Range-SAR.png Size: 24 KB

Same period, same instrument, H1:
Attached Image (click to enlarge)
Click to Enlarge Name: EURUSD-H1.png Size: 28 KB

Close-up of the trend line internally found by the inid:
Attached Image (click to enlarge)
Click to Enlarge Name: trend-line.png Size: 13 KB
No greed. No fear. Just maths.
  • #3,654
  • Edited 4:06am May 18, 2018 3:40am | Edited 4:06am
Quoting PipMeUp
{quote} Goertzel is a way to estimate the strongest sin wave to filter out. But Goertzel also lags a bit and there happens to be two cycles in force at the same time. The averaging of all the SMA weighted by the strength of the response of the Goertzel filter bank is a pragmatic way to mitigate the problem.
Got it. Thanks for the idea Pip :-)


Quoting PipMeUp
{quote} The greenish indi in the picture is the periodigram of the market at each bar. The lighter the green the higher the energy at this period (from 2 below to 80 at the top). The red line tracks the highest energy. You can see jumps where there are two big cycles.
Crumbs....impressive mate and I can see what you are trying to achieve here.

Quoting PipMeUp
{quote} Originally my idea was to detect and bracket the flat markets to play convergent strategies. The boundaries of the envelope give a definitive stop beyond which you should consider a breakout is probably happening. I never succeeded in building such a strategy. The always end up to hold two of three positions in lose at the BO time which are enough to wipe the few pips collected in the range. => ranges are too short lived.
I must admit I am a total addict towards divergence. There is more than enough offerings in this subset of price movement to keep us honest particularly if we diversify our search towards divergent opportunities. The potential open ended promise of divergence is an added string to the bow. :-)

Quoting PipMeUp
{quote} - Mann-Kendall Test tells you the probability that some monotonic trend, non necessarily linear, exists in a dataset.
- Theil-Sen estimator tells you what is the slope of a linear trend if there is one.
Thanks for the heads up on this. Interesting

Quoting PipMeUp
{quote} What about scanning the recent dataset with all windows of size say 5 to 30 in search of the best fitting linear regression channel using Mann-Kendall and Theil-Sen? The top/bottom 2 sigma give you a way to trail a stop-and-reverse. If the noise is not fat tail you can bracket the swings and build an automatically tuned pSAR. Too bad it does work out of the box... I had to add many heuristics . Also it only works in range bar charts because time charts have fat tail distributed returns.
I like this mate. Great idea. That way we are still applying discreet data intervals but taking advantage of the optimal interval selection. Nifty. In regards to fat tailed noise...our portfolio approach is a way to dampen adverse 'black swan' noise and capitalise on any 'white swan' noise through our divergent 'open ended' approach.

If we dropped the 2 sigma top bottom but rather added a top/bottom Theil-Sen on the adverse side of the trend with an envelope to allow for a degree of noise....would this possibly work as a trailing stop for the current trend and a possible stop and reverse point?....or would you advise the 2 sigma off the Theil-Sen?

Cheers Pip. This is great stuff. You are leading me on a journey into more heavy duty data analysis and indicator building and I am slowly picking things up along the way.
  • #3,655
  • May 18, 2018 6:16am May 18, 2018 6:16am
Don't confuse fat tail and black swan. A 5 ATR sudden move isn't a black swan. A black swan is an event not even considered by you model. A black swan is more like you're trading "Dutch guilder vs Deutch Mark" and suddenly you trade "EUR vs EUR".
No greed. No fear. Just maths.
  • #3,656
  • May 18, 2018 6:26am May 18, 2018 6:26am
Quoting Copernicus
If we dropped the 2 sigma top bottom but rather added a top/bottom Theil-Sen on the adverse side of the trend with an envelope to allow for a degree of noise....would this possibly work as a trailing stop for the current trend and a possible stop and reverse point?....or would you advise the 2 sigma off the Theil-Sen?
Theil-Sen is just another way to perform a linear regression. It is like SDC just bit more immune to the non-white noise. you still need to decide the width of the channel (1, 2, 3 SD). The idea of the indicator was indeed to replace the parabolic SAR for which I dislike the parameters out of the blue. It shares the same flaws. The notable difference is there are grey dots to color code the undecided/undefined trend. The dots aren't always plotted at the second stddev. Here is the heuristic part of the code.
No greed. No fear. Just maths.
  • #3,657
  • May 18, 2018 6:53am May 18, 2018 6:53am
Quoting PipMeUp
Don't confuse fat tail and black swan. A 5 ATR sudden move isn't a black swan. A black swan is an event not even considered by you model. A black swan is more like you're trading "Dutch guilder vs Deutch Mark" and suddenly you trade "EUR vs EUR".
Hehehe I like the analogy. Got it Pip. I will keep the "outlier" terminology and drop the "Black Swan" :-)

Quoting PipMeUp
Theil-Sen is just another way to perform a linear regression. It is like SDC just bit more immune to the non-white noise. you still need to decide the width of the channel (1, 2, 3 SD).
Thanks mate. Crystal clear.

Quoting PipMeUp
Here is the heuristic part of the code.
???????????
  • #3,658
  • Edited 8:22pm May 18, 2018 7:20pm | Edited 8:22pm
I am trying to kill Karma with the worst choppy non-trending markets which were very problematic for EDTT bringing losses after few years of backtesting. Cannot get under 0% with Karma. I will put some results here if I find one. Working on 1h with 4h filter as Thomas suggested.

Attached Image (click to enlarge)
Click to Enlarge Name: 8WQPxHq.png Size: 382 KB


Coffee CFD last two years:

EDTT -4.13%

Karma +4.40% risk 1% per position.
DD -9.68%

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot1.png Size: 89 KB

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot2.png Size: 198 KB
Cut short your losses. Let your profits run on. David Ricardo (1772-1823)
  • #3,659
  • May 18, 2018 8:40pm May 18, 2018 8:40pm
Quoting Ramadas
I am trying to kill Karma with the worst choppy non-trending markets which were very problematic for EDTT bringing losses after few years of backtesting. Cannot get under 0% with Karma. I will put some results here if I find one. Working on 1h with 4h filter as Thomas suggested. {image} Coffee CFD last two years: EDTT -4.13% Karma +4.40% risk 1% per position. DD -9.68% {image} {image}
Awesome R. Just fine tuning the EA with the developer so we can have lots of options and soon we will be able to fly with our analysis. Keep us posted.

So much on the plate at the moment. Down the track I also want to get into some of the stuff Pip suggested but we need to put things to bed step by step :-)

We need some music to keep life sane as all I am seeing is equity curves at the moment and I am getting cross-eyed and painless.

Inserted Video
  • #3,660
  • Edited 10:54pm May 18, 2018 9:37pm | Edited 10:54pm
Quoting Ramadas
I am trying to kill Karma with the worst choppy non-trending markets which were very problematic for EDTT bringing losses after few years of backtesting. Cannot get under 0% with Karma. I will put some results here if I find one. Working on 1h with 4h filter as Thomas suggested. {image} Coffee CFD last two years: EDTT -4.13% Karma +4.40% risk 1% per position. DD -9.68% {image} {image}
Having some joy as well using H1 as signal generator and H4 as filter on EURUSD. This was a problematic time for EURUSD with EDTT between 1 Jan 2016 to 30 April 2018

EDTT
Total return for the period -3.32%
Max Draw 8.45%

Karma Cross H1/H4
Total return for the period -1.69%
Max Draw 6.60%

So it is standing up well in comparison....but still very early days. If we have a better solution here with Karma than EDTT for our respective universes....we have a fully automated option that will give us more time at the beach for me and the alps for you :-)

Best case scenario: We find another powerful divergent system to add to the portfolio blend that offers a degree of correlation offset with robust risk weighted metrics and progressively reduces our dependence on discretionary entries;
Mid case scenario: We find a replacement for EDTT that allows us to put a 'fully systematic' stamp on our portfolio....this is actually a best case for my mindset as I hate any form of discretion.....hence the need to go to a dynamic portfolio as well as opposed to a chosen universe;
Worst case scenario: Continue on with EDTT and reverter and search for another system perhaps using Pips ideas.

Thanks to T in his help with this :-)

Here is an overlay of H1/H4 Karma on EDTT for DJ30 for the same period (smaller square) under full automation.

Attached Image


Clearly when EDTT soars.....so does Karma (so no probs there)...however we need to address the whipsaws. It looks like a filter may be the solution such as a Donchian breakout such as what we apply to EDTT to prevent trading during non-trending periods.

Once again this is steering us towards the same concluding logic....that it is best to catch the meat of the trends as opposed to worrying about capturing it's commencement. Makes sense really in the context of where serial correlation is most likely to reside in a time series or dominate the noise.
2 traders viewing now