Trading Strategies Based On Digital Filters - page 66

 
SIMBA:
K,

Thanks for posting your analysis,I will post my comments on them once I exactly understand the meaning..BTW,don`t see this as confrontational,just the opposite,I have been mistaken more times than rigth,so,the issue is not to compete,but translate a method into tradable "signals/directions",and then compare relative value.

Is your view on EURUSD H4+H1 ..Long?No cycle,trend UP?

Same for GBPUSD?

GBPJPY undecided?bouncing up and down in a triangle?

Regards

Simba

I dont see it as a confrontational.

EURUSD up to at least 1.373 than if broken higher,

GBPJPY depends if it breaks this 139 than up to 141, if not than 136

GBPUSD up

GBPJPY and GBPUSD are quite often correlated than if GBPJPY will break than a sign that GBPUSD will also go up ===> strong pound

If period is > 30 bars than can be considered as a trend so no cycles, It happens here on H4 for EURUSD and GBPUSD. Than i look for a cycles on H1.

Krzysztof

 

corona

here is more details

Krzysztof

Files:
coronacharts.pdf  298 kb
 
fajst_k:
here is more details Krzysztof

OK,

K,Ehlers usually talks about>50 periods is trend ,etc...but he usually works on D1 timeframe..so,this is like 300 H4 bars,and 1200 h1 ones...so,we shouldn`t take these 50 periods as a mantra...

Simba

 
SIMBA:
OK,

K,Ehlers usually talks about>50 periods is trend ,etc...but he usually works on D1 timeframe..so,this is like 300 H4 bars,and 1200 h1 ones...so,we shouldn`t take these 50 periods as a mantra...

Simba

Corona has a filter bank to 30 maks, it will display 30 i think when is > 30.

Forex cycles seems to be long ones i think at least on H4.

Krzysztof

 

Good call

fajst_k:
I dont see it as a confrontational.

EURUSD up to at least 1.373 than if broken higher,

GBPJPY depends if it breaks this 139 than up to 141, if not than 136

GBPUSD up

GBPJPY and GBPUSD are quite often correlated than if GBPJPY will break than a sign that GBPUSD will also go up ===> strong pound

If period is > 30 bars than can be considered as a trend so no cycles, It happens here on H4 for EURUSD and GBPUSD. Than i look for a cycles on H1.

Krzysztof

Very Good calls...

Congratulations the 3 of them were right on spot.

Regarding GBP...I usually check GBPJPY,GBPUSD and USDJPY...What I have usually found is that GBPJPY and USDJPY usually have a good level of correlation...so,usually this means that the pair to trade is usually GBPJPY...I will explain..if both GU and GJ cycles point UP..and UJ points UP too..then obviously long GJ is the best trade.

If both GU,GJ point UP and UJ points down,in theory GU is the best long,but,in practice this doesn`t happen so often since GJ and UJ tend to point in the same direction more often than not.

The same"triangle" concept is valid for shorts,or for any other pairs to use(EURGBP,GBPCHF,GBPAUD...)

I will try to post something this evening or Tuesday Wednesday..as of this moment H4 cycles are up for AJ,GJ..H1 cycles are nearing a downturn ...so,probably best moment when we have synchrony of the 2 timeframes again..but who knows?AUDJPY is at 67.71 at this moment..in the 68.25 area there is very strong resistance(6tf of January previous top and weekly m5 pivot very near),so,if it stops there I may take a chance with the H1 cycle(against the H4).

Regards

Simba

 

That's really amazing cycles operativity guys,

but I would like to make a step back to basic research.

I have started analysing GBPJPY with a self-made tool of mine ;-) . The goal is to fine tune MESA+(detrending and smoothing) with respect to two main parameters: 1. order of autocorrelation (degree) and 2. length of analysis in bars.

It is well accepted that if you take longer period of observation with MESA, you find more stable cycles.

Here follows 4 analyses of GBPJPY H4 taken for 200 subsequent bars with step 2 (one bar each two) from 2008.01.16.

The spectra are evaluated between 2 and 100 bars period, so the result is a 100x100 matrix that can be imported in a data visualization software. I have used Paraview (great free tool, btw).

You can see on the X axis the "period" variable, on the Y axis "time" (Y=0 means first bar, Y=99 means last bar or last "instant spectrum" taken) and on the Z axis the amplitude of spectrum.

The first is a 200 bars-150degree analysis. As the length is short, spectra consistently vary from one observation to the subsequent. Anyway we can see "rows" of peaks within the same areas.

As we take longer windows (400, second picture) we can see that variations are less evident.

Third picture is a 600 bar window - 150 degree observations, and fourth is a 2000-150 observation. All these are without detrending/denoising.

Next step will be increasing degree, then apply detrending and denoising.

Hope you enjoy.

Files:
 

Cycles H4

Hi,

Nice pictures. Regarding cycles H4, I believe this is a basic

https://www.mql5.com/en/forum/178842

post 89 and 90.

whatever somebody is saying about cycle stability

Regarding finding parameters value. It's not the easiest just use EA and optimizer to find optimal values of parameters money wise ??

So did you add any sort of 'frequency scanner' or whatever to have your system fully adaptive ??

How did you get so beautifull spectrum of Goertzel on this MATLAB signal

(trend with noise) ?? Did you detrend ??

Krzysztof

 

somebody

fajst_k:
Hi,

Nice pictures. Regarding cycles H4, I believe this is a basic

https://www.mql5.com/en/forum/178842

post 89 and 90.

whatever somebody is saying about cycle stability

Regarding finding parameters value. It's not the easiest just use EA and optimizer to find optimal values of parameters money wise ??

So did you add any sort of 'frequency scanner' or whatever to have your system fully adaptive ??

How did you get so beautifull spectrum of Goertzel on this MATLAB signal

(trend with noise) ?? Did you detrend ??

Krzysztof

K,

I will let Rich reply to you regarding his beautiful pics ,which,btw,seem to confirm that at H4 there is structure,but only for those that know how to look for it..while I will try to reply to your comments regarding h4,and finding optimal values..

1-You did 3 good predictions,I am not going to take into account that gbpusd didn`t go up,nor that eurusd stopped dead in its tracks and reversed,since you already explained that as a possibility,so,as I wrote before,your 3 "predictions" were good...You did those predictions using h4+h1..Now tell me,can you do any interesting prediction using m1?

2-Yes,the more bars sampled,the less error you have...WELL ONLY IF THE 2 DATASETS HAD EQUAL SIGNAL TO NOISE RATIOS PER BAR ..H4 is mostly signal,m1 is mostly noise,you are comparing pears with apples...AND,you just have enough apples at h4,so,use them...In any case I don`t care about convincing you,I just want the readers of this thread to take care,cycles below H1 have negative edge,you will lose money trading them.

3-Reality is reality,and theory is good in theory,but in practice,practice works better;)...more spaghetti for you(you lived 15 months in Italy so you must like them )A few weeks back you were obsessed with the optimal timeframe,reading CB thread,etc ..Now you are the Apostle of m1...but use H4+H1 to make your predictions..who understands you?I told you that time based the optimal timeframe is h4,all our tests with EAs show this...very simple..let`s presume best settings in h4 is 1 cycle slope from 30 to 60 periods...if we go to h1,1 cycle slope from 120 to 240 periods...it should be the same,or at least similar,isn`t it?It isn`t...or if we go to m15 it should be 1 cycle slope from 480 to 960 periods?..Again,it should be,but it isn`t...and,if you go UP to D1..from 5 to 10 days..again,it should,but it isn`t..what you will have is something like this(for the same period analyzed,let`s presume 1 January to date)and using the same cycle adapted to each timeframe...

H4=PF 5.10...H1=PF 2.30...M15=PF(0.80 to 1.61)...M5=PF (0.4 to 1.21)..D1=3.80

So,what is the optimal tf to trade?..The one that gets the best PF,Profit and %Drawdown in your tests,and ,at least with MESA and Goertzel based EAs,our tests show that the best tf is H4...Always,and second best is either D1 or H1...ALWAYS

Does this happens for all methods(Fibonacci,price action,etc)?NO...I use an israeli datamining program,WizWhy,it finds ALL patterns in the data,and I worked on pure price action patterns more than a year ago,you know extracting patterns like IF (C1>C2 & L1<L2<L3)Then BUY(Wizwhy tells you the rule`s probability and the error probability ..It is a beast)..well,I can tell you that for pure price action patterns,the best tf is D1,followed by W1.

So,the best timeframe-for your method-to trade is the one where your tests show the best results .

You know ,I am an empiricist at heart...

Your second question...regarding finding parameters,well,I think the reply is implicit in my previous words...Testing is necessary,not sufficient,because optimals do not exist in real Live trading,so,you need a "conceptual basis" that allows you to adequately frame your test results to be useful in the future..in the case of Rich,it can be,as his pics show,finding an alternate way to look at the problem to find the structure...in my case it was using MESA+SSA to find the stable cycles,then test them with EAs and confirm that they were indeed useful in practice.

Conclusion: Conceptual Frame+Extensive Tests= The Way to Go.

Regards

Simba

 
fajst_k:

So did you add any sort of 'frequency scanner' or whatever to have your system fully adaptive ??

How did you get so beautifull spectrum of Goertzel on this MATLAB signal

(trend with noise) ?? Did you detrend ??

Krzysztof

Well, I haven't yet. What I'm planning is to have goertzel + mesa working together to render the best spectrum in some LSE or other metric sense. But before I have to master how the two behave respect to detrending and denoising, that is my actual work.

The goertzel spectrum you enjoyed comes out of V1, less a little mistake ;-) plus a little reorganization.

It has a linear detrending, but i'm planning to put it on the bench with SSA/HP detrending/denoising like I'm doing now with MESA.

 

Let's see how increasing the 'degree' parameter influences the spectrum.

The following analyses are taken with length=600, same currency pair and start time, and degree value of 150, 300, 450.

As the degree increases, peakes become sharper and a peculiar thing happens: as you can see in the second picture, spectrum starts with a high peak at the beginning (around 55 period bar) and then it bifurcates in two distinct peaks.

The same bifurcation is visible for degree=450 (3rd pic).

So, the more the degree, the more the sharpness and resolution of peaks (and the more computational power needed).

But the answer is: what that bifurcation means? Does it mean that the two peaks (modes) originated from only one or that they were distinct (although near) from the beginning?

Files:
600_1_150.jpg  10 kb
600_1_300.jpg  13 kb
600_1_450.jpg  9 kb