Fractals, ZZ & the Pissing Dog | Page 20 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

ECN && STP && Scalping && Hedging - Broker Review - Oct 2009 24 replies

Alert on Fractals (MT) 2 replies

Zigzag & Fractals 10 replies

EA wanted for sound alert depending on fractals condition 2 replies

Fractals Edge Fx 3 replies

Attachments: Fractals, ZZ & the Pissing Dog
Exit Attachments

Fractals, ZZ & the Pissing Dog

  • #381
  • Edited 9:32am Feb 3, 2014 9:20am | Edited 9:32am
Last two trades AUDUSD in Trade-Explorer are testing the algo to get gracefully (?) out of an hedge.

G.
  • #382
  • Feb 3, 2014 9:48am Feb 3, 2014 9:48am
@ gg53
very good,
I have a doubt that, with respect to implementing "monty hall" in "next bar". I try to find the similarity with fractal pattern match:

1 - [Actuall bar]
2 - [actuall bar + 0]
3 - [actuall bar + 1]

actuall bar = sequence from last 12 bar

There may be "Monty Hall" a pattern of greater length of 3?
That is my question

thank you very much
  • #383
  • Feb 3, 2014 10:01am Feb 3, 2014 10:01am
Quoting burnssss
@ gg53 very good, I have a doubt that, with respect to implementing "monty hall" in "next bar". I try to find the similarity with fractal pattern match: 1 - [Actuall bar] 2 - [actuall bar + 0] 3 - [actuall bar + 1] actuall bar = sequence from last 12 bar There may be "Monty Hall" a pattern of greater length of 3? That is my question thank you very much
burnssss,

The "3 doors" or Monty Hall can be 3 or more fractal pattern length.

G.
  • #384
  • Feb 3, 2014 10:21am Feb 3, 2014 10:21am
@ gg53
then it is right that if we have a high probability of actuall bar bullish but the market gives me bearish bar, apply "monty hall" the whole sequence?
We also have to check that "next bar" not have more probability than actuall bar, but if both have high probability in one direction, as we solve this problem?

Many thanks
  • #385
  • Feb 3, 2014 3:16pm Feb 3, 2014 3:16pm
Quoting gg53
Creating SET2: From the same historical data on M5 (2011-2012) we now create our ZZ and Fractals dataset. we apply 2 standard MT4 indicators, Fractals and ZigZag (13,8,5) on that historical data. An MT4 script that writes the results to CSV will be posted later on. For EACH ZZ leg we store its fractal sequence and the total length (bar count) of that ZZ leg. UP fractal is marked as "1". Down fractal is marked as "0". So a record of each ZZ leg should look something like this: {10011} - fractal sequence within that ZZ leg 12 - total length (bar count)...
This is a really interesting work in progress.

When referring to "Total length (bar count)" are you referring to the number of candles within the zz leg?
  • #386
  • Feb 3, 2014 5:50pm Feb 3, 2014 5:50pm
Quoting mining
{quote} This is a really interesting work in progress. When referring to "Total length (bar count)" are you referring to the number of candles within the zz leg?
Yes. From beginning of zigzag fractal +1, to end fractal.

Carl
  • #387
  • Feb 3, 2014 5:58pm Feb 3, 2014 5:58pm
Quoting jimsterk
hi everyone, im still far behind of gg53's indicator because of some contradictions. the logic is : "current fractal pattern" "current length" "current +1 bar (which is the next bar) in opposite direction" right? but if you look at the attatched picture, you can see: fractal pattern ----- length [] ----- 4 (the red circle in the picture) [] ----- 1 (the blue circle in the picture) this means that the prediction is confirmed for "null fractal". i can see many "null fractal prediction" at gg53's pictures. my understanding is that if there is no fractal,...
Jim,
It has been explained that there will be turning points where fractals have not shown up yet.
Remember, we are trying to predict a turning point (fractal). So, if the previous fractal was a low fractal (signal as shown in the screenshot), you could reasonably say that a high fractal is next. So you would really be searching for a "1".
In my database a "1" can occur (high probability) at bar 6. So using -3 and +3, we go to "next bar" after 3 bars (G's database may be different).
So, in this case, a pattern match occurred for a length of 3-9 bars of this upswing.
When "next bar" predicted a down signal, an arrows shows.

Carl
  • #388
  • Feb 3, 2014 11:38pm Feb 3, 2014 11:38pm
Quoting gg53
Last two trades AUDUSD in Trade-Explorer are testing the algo to get gracefully (?) out of an hedge. G.
Ok, we got out gracefully out of the AUDUSD hedge (Trade Explorer).

G.
  • #389
  • Feb 3, 2014 11:41pm Feb 3, 2014 11:41pm
Quoting mining
{quote} This is a really interesting work in progress. When referring to "Total length (bar count)" are you referring to the number of candles within the zz leg?
mining,

Yes.

G.
  • #390
  • Feb 4, 2014 12:57am Feb 4, 2014 12:57am
Quoting crodzilla
{quote} Jim, It has been explained that there will be turning points where fractals have not shown up yet. Remember, we are trying to predict a turning point (fractal). So, if the previous fractal was a low fractal (signal as shown in the screenshot), you could reasonably say that a high fractal is next. So you would really be searching for a "1". In my database a "1" can occur (high probability) at bar 6. So using -3 and +3, we go to "next bar" after 3 bars (G's database may be different). So, in this case, a pattern match occurred for a length...
thank you carl!

so, you mean that we predict the current bar to be a fractal or not?
and current fractal pattern +"1" or +"0" to be the turning point? (depending on upleg or downleg)

or is it just an exception when we have no fractal?

if you look at gg53's post:
http://www.forexfactory.com/showthread.php?p=7205217#post7205217
you can not predict until bar 8.

so confusing...

by the way, how about the blue circle in my last post, which is zero fractal & one down bar ?
from my database, fractal "1" occurs at bar 5 (highest probability).
and i can not predict until bar 2, if i apply "max(-3 to +3) > +4" rule.
(it is the same for fractal "0", i cannot predict until bar 2)

again so confusing...
Attached Image
  • #391
  • Feb 4, 2014 1:25am Feb 4, 2014 1:25am
Quoting jimsterk
{quote} thank you carl! so, you mean that we predict the current bar to be a fractal or not? and current fractal pattern +"1" or +"0" to be the turning point? (depending on upleg or downleg) or is it just an exception when we have no fractal? if you look at gg53's post: http://www.forexfactory.com/showthread.php?p=7205217#post7205217 you can not predict until bar 8. so confusing... by the way, how about the blue circle in my last post, which is zero fractal & one down bar ? from my database, fractal "1" occurs at bar 5 (highest...
Hi Jim,

I have same point, only can do prediction from bar 2 onwards based on what discussed in this thread.
But it can be that this rule is only applicable after a confirmed first fractal, which means if no fractal and we need to add a fractal
for querying our database there is an other rule.


BR
Huub
It is not important how often you fall down, but how often you get up again
  • #392
  • Feb 4, 2014 9:10am Feb 4, 2014 9:10am
Quoting jimsterk
{quote} thank you carl! so, you mean that we predict the current bar to be a fractal or not? and current fractal pattern +"1" or +"0" to be the turning point? (depending on upleg or downleg) or is it just an exception when we have no fractal? if you look at gg53's post: http://www.forexfactory.com/showthread.php?p=7205217#post7205217 you can not predict until bar 8. so confusing... by the way, how about the blue circle in my last post, which is zero fractal & one down bar ? from my database, fractal "1" occurs at bar 5 (highest...
No, i don't mean to predict the current bar to be a fractal. We are predicting turning points (a clarification by GG53), not fractals. I don't think the post you are referring to means you have to wait till bar 8. GG53 has stated before, that fractals show up 2 bars late. So you have to guess (with a high probablility) that there will probably be a high or low fractal to show up next.

For example, what I do is... if a receive a LONG signal on the previous bar, my algorithm will search for a [1] high fractal and determine the length, even though no fractals have occurred yet (because they would be 2 bars late). In my database, high probability [1] fractals occur at bar 6. So I know I can go to my "next bar" algo from bar 3 though bar 9.

GG53's database is different, so he obviously is searching for signals earlier. I wish I had more details about the how and why his works the way it does, but I don't know.

Carl
  • #393
  • Edited 10:58am Feb 4, 2014 10:33am | Edited 10:58am
huub and carl, thank you very much!

too many different interpretations to understand, especially for non english native speaker.

but you both seems right.
and i have just found this post from ultimate truth thread.
http://www.forexfactory.com/showthread.php?p=7162626#post7162626

so if we are at upleg and current pattern is zero fractal or fractal pattern ending with "0", we have to add "1" and compare this pattern to longer pattern.
if our current fractal pattern is ending with "1", we compare this with longer pattern, right?

and how about the blue circle?
if we apply the length rule, "max(1,2,3,4) > 5" ?
gg53 have highest probability for bar 4(or less), mine is bar 5, and carl is bar 6?

or is this a kind of error prediction and so start calculation from the confirmed zigzag end?
  • #394
  • Feb 4, 2014 2:44pm Feb 4, 2014 2:44pm
Quoting crodzilla
my algorithm will search for a [1] high fractal and determine the length, even though no fractals have occurred yet (because they would be 2 bars late).
I've been doing it both ways like Carl is doing and without adding artifically the "missing" fractal. My explanation is that at any given point we are looking at 3 patterns - current, current+1 and current+0. Therefore, if current pattern is NoN, longer patterns are 1 and 0. So, patterns containing only one fractal are still used. There's one "if" of course - that is if the longer pattern (which comes up as the most probable) gets passed on to length matching. I understand G is currently using only current pattern, not longer ones, in the length match. Given the last sentence, the concept turns out to be confirming the current swing end, rather than predicting its end. Pardon if I got it wrong

I definitely agree with Jim, there are so many interpretations possible. Signal logics needs thorough discussion.

C.
  • #395
  • Feb 4, 2014 2:57pm Feb 4, 2014 2:57pm
There are many statistical interpretations but only one can be correct. The test is, the results have gg53. Doubts are many in predicting fractal.
Many thanks
  • #396
  • Feb 4, 2014 3:37pm Feb 4, 2014 3:37pm
One of my versions is starting to turn its nose into the wind, hmm, what a pleasant view
Attached Image (click to enlarge)
Click to Enlarge Name: latest.JPG Size: 91 KB
  • #397
  • Feb 5, 2014 1:54am Feb 5, 2014 1:54am
Quoting jimsterk
huub and carl, thank you very much! too many different interpretations to understand, especially for non english native speaker. but you both seems right. and i have just found this post from ultimate truth thread. http://www.forexfactory.com/showthread.php?p=7162626#post7162626 so if we are at upleg and current pattern is zero fractal or fractal pattern ending with "0", we have to add "1" and compare this pattern to longer pattern. if our current fractal pattern is ending with "1", we compare this with longer pattern, right? and how...
Hi Jim,

There are several rules for sure the most will be true but not all the time.
A possible way to find out, i'm trying just now, is to work backwards.
If you take your posted chart form GG on this page , just take a swing point. For sure you are in step 3 of the algo.
Take from that point backwards the barstring for f.i. the last 20 bars.

Test your own step 3 algo, you know for this situation, you must have an opposite bar prediction, otherwise you have to look at your algo.
Now go one bar back, do the algo again, you must now have a negative result. Repeat this step.

In parallel check changes in fractal patterns. As long no change in fractal pattern probably you must tweak your step 3 algo if wrong swingpoint.
All the time look to the rules we know, and hope to find a structure in it.
It could be a way to find out which rules when to apply


BR.


Huub
It is not important how often you fall down, but how often you get up again
  • #398
  • Feb 5, 2014 3:08am Feb 5, 2014 3:08am
Quoting candero
{quote} I've been doing it both ways like Carl is doing and without adding artifically the "missing" fractal. My explanation is that at any given point we are looking at 3 patterns - current, current+1 and current+0. Therefore, if current pattern is NoN, longer patterns are 1 and 0. So, patterns containing only one fractal are still used. There's one "if" of course - that is if the longer pattern (which comes up as the most probable) gets passed on to length matching. I understand G is currently using only current pattern, not longer ones, in the...

Hi candero,

We have the rule for the actual pattern w.r.t. length [-3....+3] and check on [+4].
I still think if you are outside this range on the longer length and and current+0 and current+1 have lower probability then the actual length of actual pattern, you have to stay with the actual pattern and still apply step 3 algo.
F.i. [101] pattern has roughly 600 occurences, the first lengths behind [-3....+3] have higher occurences then the highest occurrences in the [101x] pattern.
So I think then the only option what you have is to stay at the current pattern.

BR
Huub
It is not important how often you fall down, but how often you get up again
  • #399
  • Feb 5, 2014 2:09pm Feb 5, 2014 2:09pm
@ GG53

Are you sure this trade explorer should be part of this thread? As this EA is based on Killing zone and not on forex_GT arrow.

Kind regards,
Dennis
  • #400
  • Feb 6, 2014 7:00am Feb 6, 2014 7:00am
<p>
Quoting emonts
{quote} Hi Jim, There are several rules for sure the most will be true but not all the time. A possible way to find out, i'm trying just now, is to work backwards. If you take your posted chart form GG on this page , just take a swing point. For sure you are in step 3 of the algo. Take from that point backwards the barstring for f.i. the last 20 bars. Test your own step 3 algo, you know for this situation, you must have an opposite bar prediction, otherwise you have to look at your algo. Now go one bar back, do the algo again, you must now have...
</p>
<p>&nbsp;</p>

emonts,

That's a very good way to "backtest".
Just remember that my database is evolving due to prediction "errors" and might be a little different than your currently.

G.
0 traders viewing now