Prediction of turning points based on ZZ on fractal calc....

[joswald]

Ok, I have read the ff thread and this is my interpretation of the workings of gg's algo. Bear in mind I cannot code but I have a whipped up a spreadsheet which should help in the initial stages or the foundation on which I think it was built. So what are the steps..

- Firstly it is needed that each bar is assigned a '0' or '1' (bear/bull)
- Then the dna calculations are needed for each bar, steps 1 to 12
- Then pivot points zz or fractal points need to be identified in the dataset
- In addition the previous 4 bar dna's are required as well (sub-sequence)
- So for each pivot, there should be 5 dna sequences that need to be extracted and stored seperately.
- Then it's a matter of matching the current dna string aka current market to all the pivot strings stored, until a match is found... bingo! then ofcourse repeat on on other tf's (5,15,30,1) for confluence.

I think eurusdd then takes it one step further or deeper in regards to data mining, but in essence I think this is what gg is doing.

The biggest challenge I think is maybe extracting the pivots dna (and sub sequence dna).

I hope some of the sharp minds here are up for a challenge, looks really promising.

Cheers

[Kruspe]

Uh, this is really some serious stuff. I like it. I also follow original Ultimate truth thread on FF. But I guess I need to look at it once again and more deeply ...

Great job guys!

[joswald]

My apologies, as well as every bar being assigned a 0/1... gg is assigning each fractal within a ZigZag leg a 0/1, duh! thus the name of this thread. I think he is also calculating the fractal counts as well and combining all 3?. I still don't understand if he uses 4 step ie 0101, 01010011, 010100110101, or just steps the dna incrementally by 1 which I think would make sense as each leg is not a factor of 4.

Still thinking my way through this..

[garyfritz]

The charts Kilian19 posted are amazing -- my too-good-to-be-true alarm is ringing off the wall.

Let's assume for a moment that gg53 really CAN produce these miraculous unbelievable charts in realtime. He's not sharing the code that does it. There are "bread crumbs" that describe the process but it's a looooong path from "bread crumbs" to "working code." gg53 might not explain the full process and we might find we have holes in our crumbs.

But even if we don't manage to duplicate his results, it might be a useful exercise to get new ideas. I'll take a look at the thread and see if I can offer any insights. I probably won't be able to contribute much to the coding.

[Kilian19]

"my too-good-to-be-true alarm"

Mine is ringing as well but crodzilla is getting partially similar results as well and gg's results are not always 100% correct but have a certain error range.

When something looks to good I am very cautious as well.

Joswald. You should read about how to define dna sub sequences again.

BAR DNA(0,1) != Fractal DNA(0,1) != ZZ Fractal DNA(0,1,-1)

We work with 4 different sets of data.
I will just quote something I wrote in the thread:


Data: 0[0],0[1],0[2],0[3],1[4],0[5],1[6],0[7]

c: increment by 1 starting from 0
b: sequence length - 1
a: increment by 1 starting from 1

c + i * (b-1) < Bars hast to be true

Sequence of 3 so b = 3

for(i = 0 up to b-1)
A[c + i * a]

a = 1
0 [0], 0 [1], 0 [2] | c = 0
0 [1], 0 [2], 0 [3] | c = 1
0 [2], 0 [3], 1 [4] | c = 2
0 [3], 1 [4], 0 [5] | c = 3
1 [4], 0 [5], 1 [6] | c = 4
0 [5], 1 [6], 0 [7] | c = 5

a = 2
0 [0], 0 [2], 1 [4] | c = 0
0 [1], 0 [3], 0 [5] | c = 1
0 [2], 1 [4], 1 [6] | c = 2
0 [3], 0 [5], 0 [7] | c = 3

a = 3
0 [0], 0 [3], 1 [6]
0 [1], 1 [4], 0 [7]

Now do the same with a sequence length of b = 4,5,6,7 etc ...


With around 3000 bars you get an enormous amount of sub strings. that why it takes a few minutes to calculate the signal I don't think that you will be able to achieve that with an excel file as it gets to complex after some time.

I actually got to the very last point which is the calculation of the ZZ End point. Once I have a bit more time I can explain it little bit.

[garyfritz]

If you're looking at A[a * b + c], and you start with a = 1, b = 3, c = 1, then won't you start at A[4] instead of A[0] as you show in your first example?

Using A[a * b + c] I would assume c is the base (starting) location, a is the step size (how many to skip between selections) and b steps from 0 to length-1? So in this first line:
0 [0], 0 [1], 0 [2]
...c = 0 (start the sequence at cell 0), a = 1 (step 1 between each selection, from [0] to [1] to [2]), and b goes from 0 to 2. Then for the next line:
0 [1], 0 [2], 0 [3]
...you increase c to 1 and again sweep b from 0 to 2.

That produces cell indices that match the example you gave. Am I missing something?

With around 3000 bars you get an enormous amount of sub strings.

No kidding. Do we have any feel for how many bars are actually needed? (Apologies if it's explained in the thread, I'm just getting started...)

[Kilian19]

If you're looking at A[a * b + c]....
That produces cell indices that match the example you gave. Am I missing something?

With around 3000 bars you get an enormous amount of sub strings.

No kidding. Do we have any feel for how many bars are actually needed? (Apologies if it's explained in the thread, I'm just getting started...)

Yes sorry. I didn't think about it to much you are correct. I will fix my post in a few minutes.
We never really agreed on how many bars we take into consideration. I think it ranged from 1 Month to about 3 years and max substring with the length 13 - 20.

Currently we are down to 2 years and sequences ranging from 4 - 13.

We calculate an awful amount of information. Filter those throw half of it away and store the rest of it in files or a database so the major part of the calculation just has to take place once.

See post 523 (this one is a key post)

After some analysis (MATLAB & R), I'm currently using only last 2 years of data.

2010-2012 = training period
2013 forward testing

Let's assume for a moment that gg53 really CAN produce these miraculous unbelievable charts in realtime. He's not sharing the code that does it. There are "bread crumbs" that describe the process but it's a looooong path from "bread crumbs" to "working code." gg53 might not explain the full process and we might find we have holes in our crumbs.

I am pretty optimistic that I got 95% of it figured out. There is one little thing I am not 100% sure about but a little bit trial and error is always acceptable isn't it

[scheinwerfer]

Very interesting stuff this is.

In regard to ONLY bar sequencing - probably the best explanation is Ralome's post in the original thread. http://www.forexfactory.com/showthread. ... ost6867137
And this is all linear sequencing, some say that if A is bigger than 1, which means we are not sequencing consecutive bars, it's nonlinear. I disagree with this.

How many bars are actually needed - if we look at only bar DNA sequencing, then Eurusdd said you need all bars you can get your hands on. It has been proven in the main thread that it doesn't give any substantial benefit. I think if our aim should be to to make the fractal/ZZ thing a reality, theoretically we would need 1-2 years of data most.

Can someone tell me why do we need longer than 10-string sequences? Shorter ones have much more weight because their number is many times bigger, therefore the short strings call the shots or don't they?

[Kilian19]

Very interesting stuff this is.
Can someone tell me why do we need longer than 10-string sequences? Shorter ones have much more weight because their number is many times bigger, therefore the short strings call the shots or don't they?

Well 13 is not much longer than 10. gg made some backtests with R and determined that this gives the best results.
The longer a pattern the more unique it is and the more information it can carry. Of course there has to be a fair ratio between length and actual amount of patterns. You can look at it in the reverse order as well. If you have a string of length 1 [0] or [1] does this really tell you much about in which direction the market will go? Probably no... Why are higher timeframes more important regarding resistance and support than shorter timeframes?

[Kruspe]

Kilian, I was reading original ff thread today once more. The first 18 pages are pretty clear. I understand how to create linear subsequences etc. But now I fight a little bit how to combine it with Zig-Zag. I have to read it carefully till the end and then rather once more again. I´m not a programmer, but I know a few coding wizards (they are not SH forum members) who are able to run it on powerful high-end machines. I just have to convince them it does worth it.

I guess it shouldn´t be a problem