Fractals, ZZ & the Pissing Dog | Page 2 | 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

  • #21
  • Jan 4, 2014 12:57pm Jan 4, 2014 12:57pm
Quoting gg53
{quote} sirwolf, That's exactly what I'm trying to do right now -->> To gather the most recent info into just few posts and indexed documents. Please be patient, I don't have much spare time - so it might take a while... G.
Great GG! That is what I was hoping for. I will keep watching this space and will point my friend here when the information is ready.

Also, I am a systems analyst with a degree in computer and information sciences and 25+ years experience. I do struggle with the concepts but hope to help out. I am hoping if it is all captured in a few posts I may be able to understand and offer some help.

Looking forward to where this is going...

Bill
Believe and Achieve!
  • #22
  • Edited 4:45pm Jan 4, 2014 4:11pm | Edited 4:45pm
Sleeves rolled, let's collect the data:

We need 2 years of historical data on M5 (5 Min. chart). It is crucial to get it from a reliable source.
I'm using 2011-2012 data - leaving 2013 onward as an "out of sample" for predictions tests.

After getting the historical data, we analyse each and every bar and mark them as follows:
Down bar are marked as "0" (Close price lower than Open price).
Up bar is marked as "1" (Close price higher than Open price).
Doji bars (Close price equals Open price) are marked as the previous/preceding bar.

The end result should be a long list looking something like this {010110100010101000011001001.......} - a very long binary string.

Now we devide the long string into substrings of length 4-13: there are several ways of doing it - I'll ask others to contribute their way.
My way: you move one item to the right for each sequence length.
Example:
We have a string of 6 items, {011001} - and we want to create substrings length 4-6.
0110 - 4 length (start from first item, count 4)
1100 - 4 length (start from second item, count 4)
1001 - 4 length (start from third item, count 4)
01100 - 5 length (start from first item, count 5)
11001 - 5 length (start from second item, count 5)
011001 - 6 length (start from first item, count 6)

We store the results of our raw data and substrings in a database.
Several very good routines to communicate with the databae will be posted later on
...and a great Thank-You for that goes to http://www.forexfactory.com/go4it !!

From now on we call that one as SET1.

Creating of SET2 & SET3 in the next post.

G.
  • #23
  • Jan 4, 2014 4:46pm Jan 4, 2014 4:46pm
Attached is the DT ZZ indicator they may or may not be of help. I use it for my setups and it doesn't repaint.
Attached Image (click to enlarge)
Click to Enlarge Name: 2014-01-04_1643.png Size: 61 KB
Attached File(s)
File Type: mq4 DT_ZZ.mq4   4 KB | 2,427 downloads
  • #24
  • Jan 4, 2014 4:54pm Jan 4, 2014 4:54pm
Quoting txfxtrader
Attached is the DT ZZ indicator they may or may not be of help. I use it for my setups and it doesn't repaint. {file} {image}
txfxtrader,

Thanks.
Let us finish the introduction with all the SET1-3, logic, probabilities calculations etc. - and we will test this baby too...
Thanks again.

G.

Edit:
A second look at the thick black lines reveals there are higher-highs both before and after the indie marked top ZZ.
Nothing wrong with that, just an observation...
  • #25
  • Jan 4, 2014 7:40pm Jan 4, 2014 7:40pm
Quoting txfxtrader
Attached is the DT ZZ indicator they may or may not be of help. I use it for my setups and it doesn't repaint. {file} {image}
Hi txfxtrader,

Thanks. It is nice, and have "history" red dot of previous zzEnd (not final zzEnd). I will try it and see it live on Monday after market opened.

-smallcat
Attached Image
  • #26
  • Jan 4, 2014 8:07pm Jan 4, 2014 8:07pm
Quoting txfxtrader
Attached is the DT ZZ indicator they may or may not be of help. I use it for my setups and it doesn't repaint. {file} {image}
Quoting gg53
{quote} txfxtrader, Thanks. Let us finish the introduction with all the SET1-3, logic, probabilities calculations etc. - and we will test this baby too... Thanks again. G. Edit: A second look at the thick black lines reveals there are higher-highs both before and after the indie marked top ZZ. Nothing wrong with that, just an observation...
i check it again, and find that some times the end of ZZ is not the same as fractals.
I still use the "Custom Zigzag" by Killian.

Edit:
For alone using of ZZ, this is good. But for using with GG's idea (together with fractals), i think it must be changed a little, like Killian changed the ZZ indicator to "Custom ZZ".

Thanks
-smallcat
Attached Image
  • #27
  • Jan 5, 2014 1:57am Jan 5, 2014 1:57am
@gg Nice to see in your own thread in this new year

waiting for your more posts (set2 , set3 ,,,,,... prediction )


  • #28
  • Jan 5, 2014 2:43am Jan 5, 2014 2:43am
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) of the ZZ leg.

This is also stored in a database, and this dataset is called from now on as SET2.

G.
  • #29
  • Jan 5, 2014 3:17am Jan 5, 2014 3:17am
Creating SET3:

From the list of FRACTAL sequences obtained and stored as SET2, we now create a list of sub-sequences from each ZZ leg.
Sub-sequences of fractal patterns are made the same way as described in SET1.
We create sub-sequences of length 4-13. For fractals sequence that are longer than 13 we create only the right-most 13 fractals.

If we take the previous example of {10011} fractals sequence within a ZZ leg and apply the 4-13 length - we should get the following sub-sequences:
1001 - 4 length
0011 - 4 length
10011 - 5 length

We repeat that process for each and every ZZ leg within SET2 - leaving it intact. We don't change anything in SET2.
We are using those sub-sequences to "filter" SET1 - the raw data.
Each sub-sequence in SET1 that matches those in SET2 is stored in a database as SET3.

By applying this procedure on SET1 we actually "filtered " the raw data from noises, overshooting and other mismatches - in order to follow the logic of FRACTALS - they exist both in higher and lower patterns of the same thing.

Next post we will discuss the LOGIC to use those datasets, SET1 SET2 & SET3.

G.
  • #30
  • Jan 5, 2014 4:07am Jan 5, 2014 4:07am
Okay gg53,
I am very attentive to your post. For now, your description matches my algorithm, except that I stored it count bar (-1) and not the total count bar (considering fractals). Only I have to add the number of fractals to the number of bar (-1) without changing anything else. I will overcome this small detail.
Thank you very much for this effort.
  • #31
  • Jan 5, 2014 4:12am Jan 5, 2014 4:12am
Hi gg53,

Thank you for initiating this thread as well.
I find your approach to be very innovative, well organized and documented - in a quite scientific manner.

Looking forward with great interest to you work.

Dragos
  • #32
  • Jan 5, 2014 4:27am Jan 5, 2014 4:27am
Quoting gg53
Creating SET3: From the list of FRACTAL sequences obtained and stored as SET2, we now create a list of sub-sequences from each ZZ leg. Sub-sequences of fractal patterns are made the same way as described in SET1. We create sub-sequences of length 4-13. For fractals sequence that are longer than 13 we create only the right-most 13 fractals. If we take the previous example of {10011} fractals sequence within a ZZ leg and apply the 4-13 length - we should get the following sub-sequences: 1001 - 4 length 0011 - 4 length 10011 - 5 length We repeat that...
Hi GG,

thanks for the detailed description.
I have a question about filtering.
Suppose we have a sub-sequence 0101 2000times in SET1 and the same sub-sequence 0101 300 times in SET2, then we keep this sequence in the filtered SET3 with 2000times occurance, so same as in SET1. Am i right?
Thanks
  • #33
  • Jan 5, 2014 4:32am Jan 5, 2014 4:32am
Quoting killerno
{quote} Hi GG, thanks for the detailed description. I have a question about filtering. Suppose we have a sub-sequence 0101 2000times in SET1 and the same sub-sequence 0101 300 times in SET2, then we keep this sequence in the filtered SET3 with 2000times occurance, so same as in SET1. Am i right? Thanks
killerno,

Yes, SET3 should reflect ALL the "approved" patterns from SET2.

G.
  • #34
  • Jan 5, 2014 4:35am Jan 5, 2014 4:35am
Quoting killerno
{quote} Hi GG, thanks for the detailed description. I have a question about filtering. Suppose we have a sub-sequence 0101 2000times in SET1 and the same sub-sequence 0101 300 times in SET2, then we keep this sequence in the filtered SET3 with 2000times occurance, so same as in SET1. Am i right? Thanks
gg53, if you want to help answer any questions regarding this last you have posted, to help, just tell me. This part I have understood.
Yes Killermo, occurrences of SET1 is maintained, but the patterns have to be the same in SET1 and SET2
  • #35
  • Jan 5, 2014 5:08am Jan 5, 2014 5:08am
@ gg thanks for description with example

So far my algo doing the same way of your post

let me compare all others from now
  • #36
  • Jan 5, 2014 6:28am Jan 5, 2014 6:28am
Quoting gg53
{quote} killerno, Yes, SET3 should reflect ALL the "approved" patterns from SET2. G.
Thanks GG and burnssss.
I have the same algo routin so far too. The next posts will be the interesting part. Cant wait

regards
  • #37
  • Edited 7:11am Jan 5, 2014 6:41am | Edited 7:11am
The Logic:



Based on our datasets we construct our model.

Our model has 3 consecutive parts:

  1. Fractals prediction
  2. Total ZZ length Prediction
  3. “Next bar” in opposite direction prediction

We start with a new ZigZag leg.

Watching the incoming fractals in real time, we check against SET2 what are the probabilities of more incoming fractals for what we currently have?

If the probabilities suggest a longer pattern, we wait for more incoming bars.

If the probabilities suggest that the current fractals pattern is “right” - we move to the next step:


“Total length” probabilities – based on SET2 we calculate the probabilities of current total ZZ length being “correct” or should be longer. This is done with a tolerance of +3/-3 bars.

If not the “right” length – we wait for more bars. If total length is “right” we move to the next step:


Calculating probabilities for “Next bar in opposite direction”.

We are looking for “opposite direction” because by now we are probably near the ZZ leg end and we want “next bar” to be in opposite direction to confirm that we indeed on the turning point.

Calculating of probabilities “Next bar in opposite direction” is done against SET3, our filtered raw data.


Detailed explanation on how to calculate probabilities will come later on.


Next post will be about "adaptability & prediction errors" with this model.


G.


Edit: samallcat contributed "How to" get the ZZ data.

Posted on 1st post. Thanks, smallcat!


  • #38
  • Jan 5, 2014 6:46am Jan 5, 2014 6:46am
Thank you very much gg53,
The logic also in my algorithm. Next post will be very important to me.
  • #39
  • Jan 5, 2014 10:42am Jan 5, 2014 10:42am
Quoting gg53
The Logic: Based on our datasets we construct our model. Our model has 3 consecutive parts: Fractals prediction Total ZZ length Prediction “Next bar” in opposite direction prediction We start with a new ZigZag leg.Watching the incoming fractals in real time, we check against SET2 what are the probabilities of more incoming fractals for what we currently have? If the probabilities suggest a longer pattern, we wait for more incoming bars.If the probabilities suggest that the current fractals pattern is “right” - we move to the next step: ......
Hello GG53,

Thank you, i subscribe for this thread.
I have one question : for fractal prediction (the first step), at the beginning of new zz leg, if there is no fractal , then we try to check SET2 for the "0" fractal and "1" fractal, which one is bigger. Do we need to check the candle length exist in current zz leg , or just check the fractals ?

Without checking of candle length, how can we know that the fractals pattern is "right" ?
  • #40
  • Jan 5, 2014 10:46am Jan 5, 2014 10:46am
Quoting Davit
{quote} I am open minded person after all I am an artist and love logic and play chess just about every night.I just have hard time accepting the promise that historical data,patterns can be applied to future. There are millions of variables that are unique on every 5 min chart that can't be repeated ever again. Regards
Harmonics, Fibs, Murray Math and going down to the simple Fractal Indicator actually prove that patterns are repetitious, and as I am of the same belief that Market is Fractal in nature, if these patterns were true to the M5, couldn't they also be applied to any other TF with the same expectation?

Isn't this the whole premise behind algorithms?

That being said...

This would inevitably need be automated as a system. If applied under almost any TF, the amount of time to research the patternology and work out the probabilities would be close to impossible, manually applied.

That being said...as much of the same mind as you are in that this may only offer scratch ticket levels of probabilities...it is interesting.

Just thought I'd weigh in on the fact that there are patterns that are repetitious in nature.

And

I applaud the efforts of the outside of the box thinking. Even if this does prove to give no better odds than a coin toss...at least it was an effort.
Really...

Apply any indicator or system to any single time frame.
Take the trade and switch to another time frame.
Does the trade still make sense or does it look completely random?

Full Steam Ahead GG, can't wait to see how this comes together.
No offence meant Davit.
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
0 traders viewing now