Pattern-Finding EA | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Intresting finding or simply statistic: 5min candel 14 replies

Price pattern scanner where I choose the pattern 4 replies

Need help finding a scalping broker 12 replies

Need Help Finding a Book 4 replies

Interesting finding: EUR/JPY 1 reply

Attachments: Pattern-Finding EA
Exit Attachments

Pattern-Finding EA

  • #21
  • Oct 21, 2012 9:55pm Oct 21, 2012 9:55pm
Hi Ronald,

I have a system.

I an post here the indicators, entry/exit rules. Can you help in coding the EA ?

Thanks.
Intraday only.
  • #22
  • Oct 21, 2012 11:28pm Oct 21, 2012 11:28pm
Hi RR,

Great EA you have here,
just an idea, maybe you can add the MA type (SMA. EMA, Smoothing and LWMA) ?

Thanks
Failure leads to Success if one learns from mistakes and never give up
  • #23
  • Oct 22, 2012 7:02am Oct 22, 2012 7:02am
Hey, RR!

Instead of computing min/max values for future bars across the previous instances of the pattern, have you tried to compute a weighted average, based on the correlation factor?
  • #24
  • Oct 22, 2012 7:39am Oct 22, 2012 7:39am
Quoting Barbossa
Hey, RR!

Instead of computing min/max values for future bars across the previous instances of the pattern, have you tried to compute a weighted average, based on the correlation factor?
How do you mean?

I paint the projection based on past extremes, and an expectation that future price will be inside that projection.
  • #25
  • Oct 22, 2012 8:09am Oct 22, 2012 8:09am
It's easier to write it in pseudo-code, I hope you don't mind. What I mean is something like this:

weighted_high = 0;
weighted_low = 0;
for ( i = 0; i < NumMatches; i++ ) {
weighted_high += HH[i] * Correlation[i];
weighted_low += LL[i] * Correlation[i];
}

weighted_high /= ArrayElementsSum(Correlation);
weighted_low /= ArrayElementsSum(Correlation);

Here HH and LL arrays hold the extreme values for each sequence that is to be projected. Correlation array are the previously computed correlation factors, "ArrayElementsSum" is the value of the summed elements of the array.

This way you'll end up with something like the "high probability" high and low values for the next several bars, that maybe could be used for SL/TP levels. I'll attach a chart as well.
Attached Image (click to enlarge)
Click to Enlarge Name: test.gif Size: 23 KB
  • #26
  • Oct 22, 2012 8:18am Oct 22, 2012 8:18am
I currently have manually entered buffers that take into account the entire range.

The TP Buffer pushes the take profit target closer to the entry point, and the SL buffer pushes the stop loss farther from the entry point.

I think there needs to be some sort of heatmap showing where multiple solutions overlay.
  • #27
  • Oct 22, 2012 8:29am Oct 22, 2012 8:29am
Quoting Ronald Raygun
I currently have manually entered buffers that take into account the entire range.

The TP Buffer pushes the take profit target closer to the entry point, and the SL buffer pushes the stop loss farther from the entry point.

I think there needs to be some sort of heatmap showing where multiple solutions overlay.
A heatmap would probably be best, but it could be harder to implement. Besides, it would need a larger number of projections to produce meaningful results. This is true for my idea as well.
  • #28
  • Oct 23, 2012 8:55am Oct 23, 2012 8:55am
Awesome.. Giving it a run now.. and noting outcomes.
  • #29
  • Oct 23, 2012 10:48am Oct 23, 2012 10:48am
just curious:

if this is the projection, why did it take a sell? If the projection is supposed to be correct, seems like it's going to hit SL easily?
Guess I am confused here

http://my.jetscreenshot.com/demo/m_2...nnxv-137kb.jpg
  • #30
  • Oct 23, 2012 5:56pm Oct 23, 2012 5:56pm
Thanks for this! i have been looking for this type of ea/indicator.
i will be testing this..

RR, are there still some ea's or indicators similar to this? i want to test them as well & compare them.
  • #31
  • Oct 24, 2012 10:05pm Oct 24, 2012 10:05pm
Quoting tazmaan
just curious:

if this is the projection, why did it take a sell? If the projection is supposed to be correct, seems like it's going to hit SL easily?
Guess I am confused here

http://my.jetscreenshot.com/demo/m_2...nnxv-137kb.jpg

That projection is based on the current bar.

You would need the projection at the time the trade was taken.
  • #32
  • Oct 24, 2012 10:06pm Oct 24, 2012 10:06pm
Quoting chikanesoup
Thanks for this! i have been looking for this type of ea/indicator.
i will be testing this..

RR, are there still some ea's or indicators similar to this? i want to test them as well & compare them.

Not to my knowledge. I'm pretty sure I'm the first and for now only person to release this kind of analysis publicly.
  • #33
  • Oct 24, 2012 10:12pm Oct 24, 2012 10:12pm
Quoting Ronald Raygun
That projection is based on the current bar.

You would need the projection at the time the trade was taken.
Just on this.. it would be useful (not really needed) but useful, if we can determine how many bars back to keep on screen.. maybe a different colour?

Enjoying it however.
  • #34
  • Edited 3:23am Oct 26, 2012 2:47am | Edited 3:23am
I opened a new demo yesterday .
put the ea on 15 H1 charts and it took 10 trades (and 1 this morning)
this morning 7 winners !!! just an accident ? we will see

RR what a great work so far
Attached Image (click to enlarge)
Click to Enlarge Name: l_002.png Size: 49 KB
  • #35
  • Oct 26, 2012 8:18am Oct 26, 2012 8:18am
I think there is a similar approach. WmiFor30 is an indicator that searches for patterns in the past. WmiFor30 DTW is an expert advisor. You can use Google translate for the text.

Quoting Ronald Raygun
Not to my knowledge. I'm pretty sure I'm the first and for now only person to release this kind of analysis publicly.
  • #36
  • Oct 28, 2012 7:20am Oct 28, 2012 7:20am
Anyway to keep the past predictions on the chart? Just to help my research?
  • #37
  • Oct 29, 2012 7:44pm Oct 29, 2012 7:44pm
Been doing some looking on the higher time frames, this is a big winner (talking 4.63 times deposit winner over daily on the AUS/USD this year so far.

Issue. Only 3 trades.

Does that matter? Not a chance. Drawdown only 27% or so. (I can handle that)

I cant seem to find the sweet spot on the smaller timeframes.. I noticed 95% correlation also has a positive effect.
  • #38
  • Oct 30, 2012 2:02pm Oct 30, 2012 2:02pm
Quoting triptizine
Anyway to keep the past predictions on the chart? Just to help my research?
I played around with this a bit and converted it to an indicator so I could manually see past predictions. You might find it useful. I also added matching on negative correlations. This will help detrend the pattern predication. If you don't detrend somehow you will get a bias that is just based on how the pair has been moving over your lookback.

I also made some changes so it won't overlap the matches but since this code works in 2 passes you need to manually set your corr threshold close to your results threshold in order to avoid missing something.

The issue with this code is that the prediction is based on the extremes of movement. Seems like it only takes 2 predictions to completely override everything else. Since I am not using this to try to predict direction I have not looked into any better solutions.
Attached Image (click to enlarge)
Click to Enlarge Name: patternfinder_m15.jpg Size: 96 KB
Attached File(s)
File Type: mq4 PatternFinding.mq4   11 KB | 1,723 downloads
  • #39
  • Oct 30, 2012 11:21pm Oct 30, 2012 11:21pm
Thank you for your indicator and improvements @Paradox.

I went though the code and am wondering how you apply the negative correlation when projecting the future bars onto the chart. It looks like you are combining the results of high negatively correlating patterns on top of high positively correlating patterns,

If you reverse the predictions of the negatively correlating patterns, then you might find similarities between the highly negative and highly positive correlating sequences.

I admit the current trade entry model is flawed in the sense that it focuses on the extremes of the previous price movement. The same could be applied to trades placed in such a way that x% of trades turn a profit based on the assumption that the EA entered the trade with those particular take profits and stop losses.

I'll look to applying that revised version of the trade entry model and post it here, after I get my main Internet connection back online (I'm tethered on a smartphone)
  • #40
  • Oct 31, 2012 3:55am Oct 31, 2012 3:55am
Hi RR,

Nice EA.
I have back-tested it, the entry point seems good. However, I have difficulty to activate the trailing stop in back-test mode. Any suggestion?
0 traders viewing now