Predicting the future with Fourier transforms - page 23

New comment
 
alsu:

Listed two posts above (you can also add wavelets). Compare, for example, in this picture (sorry for the quality):


I did not see something. can you explain for those deprived of nature))

IMHO, neither Fourier nor what you wrote about will work, nobody cancelled non-stacyanarity. the whole point is in correct filtering of series before use on harmonic decomposition. and there you can decompose them with anything, and simplicity of Fourier is even possible plus, instead of naive minus because of old method. or algorithm gertzel is not good.

I understand what's about to happen - you just don't know how to cook it)))))))))))))))) and all that....

 
Trololo:


IMHO, neither Fourier nor what you wrote about will work, nobody cancelled non-stacyanarity. the whole point is in correct filtering of series before use on harmonic decomposition. and there you can decompose them with anything, and simplicity of Fourier is even possible plus, instead of naive minus because of old method. or algorithm gertzel is not good.

I realise it's about to start - you just don't know how to cook it)))))))))))))))) and all that....

No,you don'T. You are simply too lazy to search for information. The Hilbert-Huang transform was created to analyse non-stationary processes and does so successfully.
 

This indicator has the ability to build a zigzag channel and (midline channel) and thus decompose it, but so far I have no idea how to trade with all these pies

https://c.mql4.com/forum/2012/02/ZZm-fx-txt.mq4

 
alsu:

Listed two posts above (you can also add wavelets). Compared, for example, in this picture (sorry for the quality):


Generally speaking, Fourier is such a bearded method that almost everything else can be considered "more modern".

There is little information (useful) on the Internet (especially in Russian), mostly only basic, so you often have to think with your own head.

Thank you, it's beautifully done. I don't quite agree about Fourier, there are a number of modern methods: parametric (Berg for example) and non-parametric (based on eigenvectors).

 
alsu:
Hilbert-Huang transform was created for analysis of non-stationary processes and it successfully copes with it.

Once upon a time I "struggled" with Fourier. At first I did it myself in Excel, then there was a library #_lib_FFT.ex4.

It works, it gives such smooth lines:

But, due to heavy calculations, I don't use it. But I am interested in your Hilbert-Huang indicator.

Can I use it?

 
I can't use it for two main reasons: first, the indicator doesn't exist in its current form (decomposition calculation is an intermediate stage in my development, the picture was two weeks old), besides, it uses a dll, which contains a bunch of other functions ... I would not want to put all intellectual property here at once :) Better do it yourself, and you can compare the results - I had to significantly augment EMD algorithm to get a stable decomposition, I wonder what someone else would do...
 
Calculations are hard too, I'm thinking of switching to opencl, it's a trendy thing these days
 
Svinotavr:

Eduard Anatolyevich, I am curious: at what level have you "put your patience stops"? I mean, how many more weeks of your life are you ready to give to research of Fourier applicability to forex? And,, what happens after your last psychological cup of patience cracks? Will you move on to something else or leave "science" altogether, "slamming the lab door loudly"? No offence, I'm just curious.


I'm not Eduard Anatolievich, but I'm very curious... What complicated question can be seriously investigated in a few weeks? Even if you don't do the insider realization schemes, tick volumes and spectrum acceleration in parallel, it usually takes longer.

And while the Fourier transform is not really applicable to forex(and much else too), I doubt you can explain why this is the case. Explain as if the interlocutor is 5 years old, abstruse and incomprehensible explanations I've seen before.

 
alsu:
Calculations by the way are heavy too, I'm thinking of switching to opencl, it's fashionable now...

Better do it yourself, and you can compare results - I had to significantly augment EMD algorithm to get stable decomposition, I wonder what someone else would do...
I get it, but I don't want to invest in computationally heavy "stuff". Fourier was enough for me. :) If there are interesting meaningful results, let me know. Meanwhile my experience shows that it's better to follow simple ways. For example, you may use a very simple indicator of your own development and have the robustness for a couple of years ahead. But even that requires a lot of resources to make it work... :)
 
Trololo: it's time to clean up your posts.
Yeah, from the very first post after death.
1...161718192021222324252627282930...55
New comment