Optimized Trend Trading | Page 29 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Optimized Risk vs Reward Equation 335 replies

Loss Less Optimized Strategy 63 replies

Optimized Parameters in Metatrader Testing? 3 replies

"Optimized Parameter" feature in EA? 0 replies

Strategy with Optimized parameter 1 reply

Attachments: Optimized Trend Trading
Exit Attachments

Optimized Trend Trading

  • #561
  • Nov 5, 2009 1:30pm Nov 5, 2009 1:30pm
Quoting MadCow
That post was quite a bit later than TopCat description. He did not have the Goertzel algorithm when he first described topcat. Some later posts discuss the coding of the algorithm.

He is defining it this way because he thinks the dominant cycle is the trend, everything else is noise. But using this definition and the Goertzel, you will not get the rapid response shown in the original topcat image because a Goertzel, or any other DFT will not change rapidly enough, since it is an average over a long block. I think his later systems may use the...
Hi MadCow,
I completely agree with you, it's exactly what I think.
BTW, a clutter filter isn't really equals to SnR indicator, for me it looks more a trend/nontrend indicator, anyone have a source code for this ?
  • #562
  • Nov 5, 2009 1:33pm Nov 5, 2009 1:33pm
Quoting MadCow
That post was quite a bit later than TopCat description. He did not have the Goertzel algorithm when he first described topcat. Some later posts discuss the coding of the algorithm.

He is defining it this way because he thinks the dominant cycle is the trend, everything else is noise. But using this definition and the Goertzel, you will not get the rapid response shown in the original topcat image because a Goertzel, or any other DFT will not change rapidly enough, since it is an average over a...
Looking back, I can see that you're probably correct. It seems his clutter filter kicks in when his main filter and range-gate filter are approaching each other in value. Could be something as simple as decreasing/increasing separation of the output of the two signal filters for n number of bars.
  • #563
  • Nov 5, 2009 3:03pm Nov 5, 2009 3:03pm
Quoting asm8086
Hi MadCow,
I completely agree with you, it's exactly what I think.
BTW, a clutter filter isn't really equals to SnR indicator, for me it looks more a trend/nontrend indicator, anyone have a source code for this ?
Hi ASM8086,

You might try the fractal dimension indicator. There are a few around. I think the latest is FGDI attached.

The problem with this indicator is that it requires a large block of data to do a reasonable job, so it's slow. You might try to use it on a large block of a small TF, say 5min bars, to decide about H1 bars, but that's problematical theoretically, if you believe CB that some TF's are better than others. CB implies that the fractal dimension of one TF may differ from that of another?? But if the dynamic process behind the price series is fractal, then does the strange attractor change from TF to TF? I thought that fractal meant that the series essentially "looks" the same no matter the magnification, and that the fractal dimension would therefore not change with TF.

Attached File(s)
File Type: mq4 FGDI.mq4   19 KB | 940 downloads


A good explanation of this indicator can be found at http://fractalfinance.blogspot.com/
  • #564
  • Nov 5, 2009 4:15pm Nov 5, 2009 4:15pm
In theory yes,
all time frames are supposed to be self-similar,
but plot a couple of histograms and you well see that the lepto-kurtosis (heavy tails of the probability distribution) increases as the time frame decreases. 1 minute data will have much more extreme observations than yearly data. All timeframes have certain properties, like Hurst exponents, autocorrelation, ...

The fact that the H1 mentioned in this thread is favored by some -including me- is associated with a certain study of a major investment bank from a couple of years ago on Mutual Information and timeframes showing that it does not make any sense to trade a time frame smaller than 60 minutes.

In my opinion, the respective DSP techniques are obviously not needed to build a clutter filter. Also, FDI is kind of useless for the purpose of filtering trend and range if you think about it. It simply says something about the general tendancy/persistence of trend in the timeseries as a whole, it's not used to filter trend and range (of the past 10 or so bars)..
  • #565
  • Nov 5, 2009 6:02pm Nov 5, 2009 6:02pm
Quoting MisterH
The fact that the H1 mentioned in this thread is favored by some -including me- is associated with a certain study of a major investment bank from a couple of years ago on Mutual Information and timeframes showing that it does not make any sense to trade a time frame smaller than 60 minutes.
Hmmm...do you have a link to this study or a name. I would like to see how they came to this conclusion.
  • #566
  • Nov 5, 2009 7:14pm Nov 5, 2009 7:14pm
Quoting MisterH
In theory yes,
all time frames are supposed to be self-similar,
but plot a couple of histograms and you well see that the lepto-kurtosis (heavy tails of the probability distribution) increases as the time frame decreases. 1 minute data will have much more extreme observations than yearly data. All timeframes have certain properties, like Hurst exponents, autocorrelation, ...

The fact that the H1 mentioned in this thread is favored by some -including me- is associated with a certain study of a major investment bank from a couple of years ago on...
Mister H..
I agree with all you say. Calculating the FDI is an averaging process, and must lag. Nevertheless it seems possible that if one estimated e.g. that the general tendency of M1 bars was antipersistent say for 120-180 M1 bars, and the general tendency of M5 bars was antipersistent, maybe for 24-36 M5 bars, then he might guess that the current H1 bar was antipersistent also. This would tip the likelihood of a successful trend trade in a negative direction.

I must admit that I don't think this would be a very effective 'clutter filter', but you never know until you try.
Regards, MC
  • #567
  • Nov 6, 2009 1:42am Nov 6, 2009 1:42am
Quoting BlackMage
Hmmm...do you have a link to this study or a name. I would like to see how they came to this conclusion.
As far as I remember UBS did some search on the subject,I dont have the link .
  • #568
  • Nov 6, 2009 1:43am Nov 6, 2009 1:43am
Quoting BlackMage
Hmmm...do you have a link to this study or a name. I would like to see how they came to this conclusion.
Seconded, I really want to read that study.
  • #569
  • Edited 2:58am Nov 6, 2009 2:19am | Edited 2:58am
I think I'm going little out of argument, but due to the presence of great mathe skill traders here, I want to post a link for a chaos kit that I have try, but due to my little knowledge of programming i wasn't able to implement the kit to make a good interface for trading, maybe for someone could be interesting.

These are the links http://www.scientio.com/products/ChaosKit
http://chaoskitinterface.codeplex.com/

Lubo.
If I know how, I tell You!! Few but Good!!
  • #570
  • Edited 4:32am Nov 6, 2009 3:46am | Edited 4:32am
Actually, H1 data have shown persistent trends (H = +/- 0.6), indicating that trend following strategies should outperform mean-reversion trading systems.
I don't have that UBS study on AMI & time frames and I forgot its title; sorry.

I think that choppiness index gives a good idea on what a decent clutter filter could be: I think ATR, high and low are good ingredients, so I'm off cooking..
  • #571
  • Edited 5:39am Nov 6, 2009 5:29am | Edited 5:39am
Hi all,

To get a coherent system his clutter filter's lag must be equals to his main filter's lag. So, perhaps he used some basic Trend/Range indicator and then he tweaked it with his ullma, I'm actually searching in this way...

BTW, to compare my results with his chart I need IG Market historical data (I actually use FxPro Metatrader's data wich has some differences with CB's data, it's why you can already see some decimal points differences between my clone and the original chart). If someone has these data, please send me.

Thanks,
Philippe
  • #572
  • Nov 6, 2009 5:43am Nov 6, 2009 5:43am
Here you have quick test of chopiness index performance. I applied Ehler's DSS strategy and DSS + chopines index to the same data set to 5 currency pairs.
Overall it improved yearly return from -153% to - 77.6% and PF from 0.72 to 0.91. In both cases performance was out of sample. Both strategies were optimized by GO.


Krzysztof
Attached Image(s) (click to enlarge)
Click to Enlarge Name: cc.JPG Size: 72 KB
Click to Enlarge Name: cc1.JPG Size: 79 KB
Click to Enlarge Name: cc2.JPG Size: 78 KB
Click to Enlarge Name: cc3.JPG Size: 57 KB
  • #573
  • Nov 6, 2009 7:56am Nov 6, 2009 7:56am
Quoting asm8086
Hi all,

To get a coherent system his clutter filter's lag must be equals to his main filter's lag. So, perhaps he used some basic Trend/Range indicator and then he tweaked it with his ullma, I'm actually searching in this way...

BTW, to compare my results with his chart I need IG Market historical data (I actually use FxPro Metatrader's data wich has some differences with CB's data, it's why you can already see some decimal points differences between my clone and the original chart). If someone has these data, please send me.

Thanks,
Philippe...

I have seen CB's ullma(I am not going to share it), it is smoothed sine weighed SMA.I have seen better filter from İgorad he would not share it though.I think it was a pole smoother but his own version
  • #574
  • Edited 9:51am Nov 6, 2009 9:38am | Edited 9:51am
Quoting purley
I have seen CB's ullma(I am not going to share it), it is smoothed sine weighed SMA.I have seen better filter from İgorad he would not share it though.I think it was a pole smoother but his own version
Purley, what is a "smoothed sine weighed SMA" or a "pole smoother" ? And what do you want to say with "I have seen CB's ullma(I am not going to share it)" ?

CB's MAs are simply Sinc (cardinal sine) Lowpass filters. I made a matlab version of Igorad's NonLagMA so we can compare the responses of these filters :

1. The magnitude respones :

Attached Image (click to enlarge)
Click to Enlarge Name: mag.gif Size: 19 KB


2. The phase response :

Attached Image (click to enlarge)
Click to Enlarge Name: pha.gif Size: 22 KB


3. The Lag response :

Attached Image (click to enlarge)
Click to Enlarge Name: lag.gif Size: 19 KB


We notice easily that CB's ULLMA 39 has a little much lag than Igorad NonLagMA 79 because CB's high frequency attenuation is really better, so if you search for a good brickwall filter use CB's version.
  • #575
  • Nov 6, 2009 9:38am Nov 6, 2009 9:38am
Here is another trend dedector

Trendiness Indicator
A common way to define the strength of the market trend is to divide the absolute value of the change in price by the average true range over the past N bars. This tells us how much the market has moved relative to the average true range. A market that's moved three times the average true range over the last month has a stronger trend, for example, than one that's moved only 1.5 times the average true range.

In EasyLanguage, this can be written as follows:

Input: NBars (30);

Value1 = AbsValue(C - C[NBars])/Average(TrueRange, NBars);
Plot1(Value1, "Trendiness");
  • #576
  • Nov 6, 2009 9:47am Nov 6, 2009 9:47am
Thanks fajst_k ! I test it right now.
  • #577
  • Nov 6, 2009 10:36am Nov 6, 2009 10:36am
For those interested in Julius Bartels' paper "Random Fluctuations, Persistence, and Quasi-Persistence in Geophysical and Cosmical Periodicities" it can be found at http://www.leif.org/research/Bartels1935.pdf.
  • #578
  • Nov 6, 2009 10:37am Nov 6, 2009 10:37am
Quoting asm8086
Thanks fajst_k ! I test it right now.

You are welcome. Than I made another test DSS + Trendiness indicator.
I smooth original one with 4 bar WMA as momentum function used in indi makes
it's very 'noisy'. See screenschoots. So overall

DSS PF 0,72 %profitable 31.65%
DSS + chopiness PF 0.91 % profitable 33.2%
DSS + trendiness PF 077% %profitable 59.6%

All test I applied to the same data set and they are out of sample. So I think
Trendiness is a winner here - it increased %profitable significiant.

Let's se what results you will get.

Krzysztof
Attached Image(s) (click to enlarge)
Click to Enlarge Name: td1.JPG Size: 93 KB
Click to Enlarge Name: td2.JPG Size: 79 KB
Click to Enlarge Name: td3.JPG Size: 62 KB
  • #579
  • Nov 6, 2009 4:13pm Nov 6, 2009 4:13pm
Apparently, Bartels' test is associated with the concept of "harmonic dial analysis". Still trying to find a clear algorithmic expository of both the "Bartels' test" and "harmonic dial analysis".

AugustLeo
  • #580
  • Nov 7, 2009 6:22am Nov 7, 2009 6:22am
Forget about Bartels. It is of no use in this context.
1 trader viewing now