Fractals, ZZ & the Pissing Dog | Page 15 | 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

  • #281
  • Jan 21, 2014 5:19am Jan 21, 2014 5:19am
killerno,

Quoting killerno
my Set's for 2011-2012:
SET1: 16368
SET2: 7977
SET3: 2114
Do you calculate all possible (4-13) fractal subsequences of SET2 to obtain SET3?
In post #29 (http://www.forexfactory.com/showthread.php?p=7189951#post7189951) gg53 says:
Quoting gg53
For fractals sequence that are longer than 13 we create only the right-most 13 fractals.
If I consider just the rightmost fractal sequence of each leg, I get ~1300 in SET3 otherwhise ~2150.
  • #282
  • Jan 21, 2014 7:19am Jan 21, 2014 7:19am
Quoting killerno
Hi, my Set's for 2011-2012: SET1: 16368 SET2: 7977 SET3: 2114 Carl, Coud you please review your probability doc file. The total length calculation is different in your doc compare how GG is calculating. (http://www.forexfactory.com/showthread.php?p=7206574#post7206574 ). Maybe we should go through the pdf and ask GG to confirm the steps. GG If you have time could you please post a newer 5min TF screenshot. Thanks for all! regards
Hi GG & Carl,

Could you please tell me how much is your SET2 & SET3 for year 2011-2012, EU M5 ?
My results below is right or wrong ?

SET1 = 16368
SET2 = 2587 (as it is)
SET3 = 1562

Thank you.
  • #283
  • Jan 21, 2014 7:48am Jan 21, 2014 7:48am
Quoting flx23
killerno, {quote} Do you calculate all possible (4-13) fractal subsequences of SET2 to obtain SET3? In post #29 (http://www.forexfactory.com/showthread.php?p=7189951#post7189951) gg53 says: {quote} If I consider just the rightmost fractal sequence of each leg, I get ~1300 in SET3 otherwhise ~2150.
Hi flx23,

I am creating only the right most fractals, same as GG described. It seems everyone get different result.
regards
  • #284
  • Jan 21, 2014 7:59am Jan 21, 2014 7:59am
@ gg53
I have used one of your first charts of "the ultimate truth".
I check that the calculation of "next bar" is right.
I find the following:

1 - precise signals using the signal bars from previous to current bar.
2 - Sometimes, a pattern, for example:

------- 1 0 0 1 213
------- 1 0 0 0 0

Removing this type of results you get good signals.

3 - The study of "next bar" looks different once was a zzleg ends.

You can tell us something?
All get the same results, the databases seem to agree.
A hint of what we do wrong?.

thank you very much
  • #285
  • Jan 21, 2014 10:47am Jan 21, 2014 10:47am
Quoting burnssss
@ gg53 I have used one of your first charts of "the ultimate truth". I check that the calculation of "next bar" is right. I find the following: 1 - precise signals using the signal bars from previous to current bar. 2 - Sometimes, a pattern, for example: ------- 1 0 0 1 213 ------- 1 0 0 0 0 Removing this type of results you get good signals. 3 - The study of "next bar" looks different once was a zzleg ends. You can tell us something? All get the same results, the databases seem to agree. A hint of what we do wrong?. thank you very much
burnssss,

I don't know what you or others are doing "wrong", if at all.
Assuming we all start with the same historical data and same SET1, SET2, SET3, my database is changing when:

1. I get an "ERROR" prediction - new elements are inserted and the SETs are re-calculated.
That makes my database different from what we all started with (assuming all started with same DB).
Those changes are inserted continously, from the begining of 2013 (out of sample).

2. The algo behaves differently when unexpected fractal arrives, in real time, before the predicted swing point is marked.
That issue will be the next one to be explained (the "3 door" issue).

G.
  • #286
  • Jan 21, 2014 10:59am Jan 21, 2014 10:59am
Quoting gg53
{quote} burnssss, I don't know what you or others are doing "wrong", if at all. Assuming we all start with the same historical data and same SET1, SET2, SET3, my database is changing when: 1. I get an "ERROR" prediction - new elements are inserted and the SETs are re-calculated. That makes my database different from what we all started with (assuming all started with same DB). 2. The algo behaves differently when unexpected fractal arrives, in real time, before the predicted swing point is marked. That issue will be the next one to be explained...
Thank you very much for your replay gg53.
Some zzleg with which I worked, the fractal patterns were expected, yet the result was not good.
My question is about calculating probability "next bar" and fractal pattern match.

1- probability of "next bar"is calculated using the bars since previous signal until actual bar?.
"next bar" and "trendy bar" both should be 100% in the same direction?

2 - Fractal Pattern match, I think for some reason, perhaps we are failing in this section. But I need to rule out the option of "next bar". I think the pattern match calculation should be correct according to the pdf, but I have an error and do not know where you can be.

I am clear that adaptability plays a very important role in your database, but that does not mean that I have right signals. I start from the assumption that my signals should be correct. To me, a good signal is that you get benefit.

Many thanks
  • #287
  • Jan 21, 2014 11:58am Jan 21, 2014 11:58am
Quoting burnssss
{quote} Thank you very much for your replay gg53. Some zzleg with which I worked, the fractal patterns were expected, yet the result was not good. My question is about calculating probability "next bar" and fractal pattern match. 1- probability of "next bar"is calculated using the bars since previous signal until actual bar?. "next bar" and "trendy bar" both should be 100% in the same direction? 2 - Fractal Pattern match, I think for some reason, perhaps we are failing in this section. But I need to rule out the option of "next bar". I think the...
burnssss,

Please try to consult crodzila regarding "next bar" and fractals probability calculation.
He did a great job assembling and compiling all in one coherent Doc.

G.
  • #288
  • Jan 21, 2014 12:04pm Jan 21, 2014 12:04pm
Quoting gg53
{quote} burnssss, Please try to consult crodzila regarding "next bar" and fractals probability calculation. He did a great job assembling and compiling all in one coherent Doc. G.
I have consulted, and the results are not as expected. There is something that is not explained in the pdf makes good signs.
I wonder how do those two steps I asked above.
Just one example is enough.
thank you very much
  • #289
  • Jan 22, 2014 12:48am Jan 22, 2014 12:48am
Quoting killerno
{quote} Hi GG, I've got my signals similar place then yours, but as soon as a long leg comes (after the 3rd down signal in your picure) I get my UP signal too early and after it keep failing. I dont know how to filter these early signals. MAybe the Handling Prediction errors what I am missing. regards
Despite not addressed to me some remarks in the hope it will trigger us.
Alternating [1010...1010] patterns (fractals or bars) have in our set2 and set3 higher
occurrences as not pure alternating patterns {100...10111]
So if looking for opposite bar in the way Carl described in his document and when at the end you
have several 000 or 111 bars you will "always" get an opposite bar for next bar and actual bar.
But in reality you are going strongly up or down. So wrong signal.
I checked actual bar prediction to actual bar value(this one we know) and if opposite I wait for next bar.
Results are soso. It gives better result for above descibed patterns, but in other occasions it does not makes it better.

Taking Carls descibtion as a lead it can mean IMO only that;
We are to quickly in step 3 of the algo, so step 1 and 2 of Carls document need some more elaboration.
or next bar prediction is not how it has to be done.
It seems an open door.

Carl I'm certainly not criticising your document, you one of the few that puts his way of working on paper.
By this we can discuss things more clearly, but for sure there must be something more.


BR Huub
It is not important how often you fall down, but how often you get up again
  • #290
  • Jan 22, 2014 9:21am Jan 22, 2014 9:21am
Quoting emonts
{quote} Despite not addressed to me some remarks in the hope it will trigger us. Alternating [1010...1010] patterns (fractals or bars) have in our set2 and set3 higher occurrences as not pure alternating patterns {100...10111] So if looking for opposite bar in the way Carl described in his document and when at the end you have several 000 or 111 bars you will "always" get an opposite bar for next bar and actual bar. But in reality you are going strongly up or down. So wrong signal. I checked actual bar prediction to actual bar value(this one we...
emonts,

I don't understand your statement about alternating fractals.
The last fractal have not yet revealed itself, it will reveal itself two bars too late.
So we have to look for current fractal pattern, current total length and wait for a predicted bar in opposite direction.
If, for the above scenario, we are accurate - the fractal will appear only after two more bars.

G.
  • #291
  • Jan 22, 2014 9:53am Jan 22, 2014 9:53am
Quoting gg53
{quote} emonts, I don't understand your statement about alternating fractals. The last fractal have not yet revealed itself, it will reveal itself two bars too late. So we have to look for current fractal pattern, current total length and wait for a predicted bar in opposite direction. If, for the above scenario, we are accurate - the fractal will appear only after two more bars. G.
Hi GG53,

I think it is some misunderstanding, maybe i did not explain good enough
I was reacting on the situation that when using Carls describtion for prediction opposite bar and the last three actual bars are f.i. all ...000 or all ...111,
then pattern like ...0001 and ...1110 have always higher occurrences then ...0000 and ...1111. For fractal patterns this also the case , but has nothing to do with prediction opposite bar.
So when the market goes strongly up and you have several 1 bars and use only Carls describtion, you will get easy opposite bar. So IMO there must be other rules too.


BR

Huub
It is not important how often you fall down, but how often you get up again
  • #292
  • Jan 22, 2014 11:33am Jan 22, 2014 11:33am
Quoting emonts
{quote} Hi GG53, I think it is some misunderstanding, maybe i did not explain good enough I was reacting on the situation that when using Carls describtion for prediction opposite bar and the last three actual bars are f.i. all ...000 or all ...111, then pattern like ...0001 and ...1110 have always higher occurrences then ...0000 and ...1111. For fractal patterns this also the case , but has nothing to do with prediction opposite bar. So when the market goes strongly up and you have several 1 bars and use only Carls describtion, you will get easy...
Hi,

Yes I have exactly the same problem. I backtested the same day as GG 17.01.14 and the first signals are "OK" almost the same, but after the bigger down move I get UP signals(there were 4 or 5 down candles in a row) and unfortunately did not get soon after a DOWN signal and it messed up the signals after. When I check GGs chart he did not get any UP signal only at the end of the move. So we are missing something definitely .
reagrds
Attached Image (click to enlarge)
Click to Enlarge Name: error.png Size: 38 KB
  • #293
  • Jan 22, 2014 12:55pm Jan 22, 2014 12:55pm
gg53,
How are you doing?
I was wondering if you have any updates on how you are planning to incorporate zigzag height into your program.

Also, do you still use R as your research tool of choice? I ask because I have learned R and I also use it, and the reason I learned it is because you mentioned that you used it ,along with matlab

R has also been immensely useful and helpful in my research.
  • #294
  • Jan 22, 2014 4:05pm Jan 22, 2014 4:05pm
Quoting killerno
{quote} Hi, Yes I have exactly the same problem. I backtested the same day as GG 17.01.14 and the first signals are "OK" almost the same, but after the bigger down move I get UP signals(there were 4 or 5 down candles in a row) and unfortunately did not get soon after a DOWN signal and it messed up the signals after. When I check GGs chart he did not get any UP signal only at the end of the move. So we are missing something definitely . reagrds {image}
killerno,

In your (mine) posted example the down signal at the start of the big move was an "error".
In those cases the algo start calculating again from the actual ZZ top after it is confirmed.
That's why it "skip" the intermediate up moves and targeted the lower swing point.

G.
  • #295
  • Jan 22, 2014 4:08pm Jan 22, 2014 4:08pm
Quoting Starry
gg53, How are you doing? I was wondering if you have any updates on how you are planning to incorporate zigzag height into your program. Also, do you still use R as your research tool of choice? I ask because I have learned R and I also use it, and the reason I learned it is because you mentioned that you used it ,along with matlab R has also been immensely useful and helpful in my research.
Starry,

Currently I'm not using it - but it was very usefull at the begining of development.
I'm currently busy implementing the same ideas on Renko charts - it seems like much more pips in every ZZ leg.

G.
  • #296
  • Jan 22, 2014 5:18pm Jan 22, 2014 5:18pm
Quoting gg53
{quote} killerno, In your (mine) posted example the down signal at the start of the big move was an "error". In those cases the algo start calculating again from the actual ZZ top after it is confirmed. That's why it "skip" the intermediate up moves and targeted the lower swing point. G.
Hi GG,

thank you for the answer. So it means that you calculating from the red circle the ZZ leg (after the red down fractal confirmed), because your signal came to early. Am I right?
I backtested that way as well and I get my UP signal, where the green arrow is, so it is still early. Also between the red circle and your UP signal there are 11 fractals, my database did not give applicable prediciton for 11 fractals. Did you used your error handling method during the calculation?
thank you

regards
Attached Image (click to enlarge)
Click to Enlarge Name: error.png Size: 24 KB
  • #297
  • Jan 22, 2014 6:15pm Jan 22, 2014 6:15pm
Quoting gg53
{quote} Starry, Currently I'm not using it - but it was very usefull at the begining of development. I'm currently busy implementing the same ideas on Renko charts - it seems like much more pips in every ZZ leg. G.

ok, thanks, that should be interesting
  • #298
  • Jan 22, 2014 6:53pm Jan 22, 2014 6:53pm
Quoting killerno
{quote} my database did not give applicable prediciton for 11 fractals.
That describes bang on what I'm having also!

Killerno, from what point do you consider the red point valid? I'm thinking now maybe the confirmation gets delayed, that would explain the signals. 3 hours without signals would seem like a malfunction (I know it isn't ).

C.
  • #299
  • Jan 23, 2014 2:31am Jan 23, 2014 2:31am
Quoting candero
{quote} That describes bang on what I'm having also! Killerno, from what point do you consider the red point valid? I'm thinking now maybe the confirmation gets delayed, that would explain the signals. 3 hours without signals would seem like a malfunction (I know it isn't ). C.
Hi candero,

For me the red "circle" ZZ is valid 5 bars after it when the low fractal after it confirmed.(you can see in the picture). I don't know what we are missing.

regards
  • #300
  • Jan 23, 2014 10:13am Jan 23, 2014 10:13am
Quoting killerno
{quote} Hi candero, For me the red "circle" ZZ is valid 5 bars after it when the low fractal after it confirmed.(you can see in the picture). I don't know what we are missing. regards
killerno & candero,

It's obvious that your algo isn't processing the "error" prediction as it should.
Also, part of that process is to insert that error zz leg into the database and recalculate.
And therefor I don't know how many such data items you failed to insert into the database during 2013.
Maybe those are the reasons why you are getting such results - or maybe still missing explanation about what to do in error predictions that occur before the predicted swing point.
So - I think that there is too much varibles missing from your side that you can't realy analize one ZZ leg against my result.
Furthermore - if you compare your results against each other - I'm sure you'll find differences...

G.
0 traders viewing now