Help with Fourier - page 12
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
You have to know how to use the PF in different ways.
Using it for purposes other than its intended use. I.e. the implications of using PF in dynamics.
I got a real spectrum filter. It automatically cuts off parasitic harmonics.
I'm surprised at the result myself. I managed to turn a disadvantage of PF into an advantage.
1. PF has a lot of different uses. What do you mean by direct assignment (or indirect assignment), let's say two-dimensional PF?
2. only a decisive rule (usually called thresholding) can cut off the spurious harmonics of the "real spectrum". Or your notion of "real spectrum" is different from the commonly known interpretation.
Here is a quote from wikipedia "The discrete Fourier transform is a special case (and sometimes used for approximation)"
What an old thread this is!
Good thing I didn't read it before. Otherwise I wouldn't have gotten into it. It's good to be an amateur on any subject. No barriers, no preconceived notions.
PF is not suitable for prediction in a static application. This is clear as it is.
No one has raised the problem of parasitic harmonics arising from price differences at the ends of the sample.
It's a 90 degree angle!!! There are all the harmonics that exist in nature on such a front!
And almost no one has used, except klot, PF in dynamics.
I made a visualizer too. And I got an amazing result.
All that remains is to write a predictor. Of course, it won't predict far from it. But the result will be almost absolute within half of the sample.
When I get the final result I will definitely publish it. And it doesn't matter what it will be. A negative result is also a result.
So how are the results coming along, will you share?
The trend can be separated. But Fourier has one disadvantage, I've already written about it above. We take a fixed interval and in order to perform the transformation we multiply this interval in both directions to infinity, as a result we have a continuous signal (rate) in infinity time, because sine waves are continuous. Example, our price slice is 10, 11, 12, 13, 12, to do the conversion we need to make a continuous series out of it ... 10, 11, 12, 13, 12, [10, 11, 12, 13, 12], 10, 11, 12, 13, 12, ... The result, the future price is clearly known, it's 10, that's why Fourier doesn't work. To apply the idea of frequencies we need to find another decomposition method. For example, you can clearly set a few frequencies and by enumeration, minimizing the error, select for them the values of amplitudes and phases, we will get a trend, but for this you need a very powerful computer.
The interpretation is slightly different. If we decompose a segment of a function into a Fourier series - a set of harmonics, then if we then sum these harmonics - we get a slice of our original function, multiplied both ways to infinity.
If we take a sample of 1024 bars, Fourier considers that 1024 bars is the period of the first harmonic.
If there are any waves with a period of 256 bars in this 1024 bar sample then the 4th harmonic will be drawn in the spectrum. If we cut a 512-bar chunk out of our sample and do another Fourier transform, we will see these waves in the spectrum as the 2nd harmonic. Etc.
If our sample contains a trend component, i.e. the final price is not equal to the initial one, the Fourier transform will try to depict this trend oblique straight line using a set of harmonics (!)
and a bunch of harmonics will appear in the spectrum that do not correspond to any waves on the chart. Therefore, if the task is to extract some periodic components from the price chart, the trend component should be removed before transformation.
Edit. We can subtract the low-frequency wave instead of the trend one, i.e. we can remove low frequencies.
The same applies to price jumps, for instance, due to news events etc.
So far the results are encouraging. >> There's still a lot of work to be done.
For the sake of interest and truth, take an integrated random variable and apply your method to it, and if the results are encouraging, you can trash everything you've worked up. If the results are inconclusive, then feel free to share your work with us! Below is a file with CB attached.
Check it out.
Here's the example (indicator) I used to study Fourier...
Look in the code there - it's not hard.
Looked it over, tweaked a few things. On the test function, it works.
For the sake of interest and truth, take an integrated random variable and apply your method to it, and if the results are encouraging, you can trash everything you've worked up. If the results are inconclusive, then feel free to share your work with us! Below is a file with CB attached.
Check the method for lice.
Sergey, do you seriously believe that the random process will always and irrevocably be random? Are you a fan of scientific dogma?
Try to consider a random two or three dimensional process in the fourth, fifth or more dimensions. It's not random there at all.
The method I have invented theoretically allows one to reduce any random process to a regular one. But it is almost impossible to apply it practically. Computer performance is lacking.
================
Unfortunately, I stopped work on this topic for a year. Now I am doing it again. I'll be sure to post a picture of what I've got.
Sergei, are you seriously suggesting that a random process will always and irrevocably be random? Are you a fan of scientific dogma?
Yes. I'm sure. That's why it's called random. Otherwise, we should be talking about a quasi-random process, etc.
Try to consider a random two or three dimensional process in fourth, fifth and more dimensions. It's not random there at all.
This method of revealing of latent regularities (informational dimension of BP), is applicable to quasi-random processes, on truly random the dimensionality of the method coincides with the dimensionality of the analyzer space. If to analyse the series I posted by this and other methods of estimation, you will be convinced in its random nature.
The method that I have invented theoretically allows you to reduce any random process to a regular one.
Zhunko, you have to be modest and cautious here. Modesty is just an embellishment, while caution, allows you not to kick back publicly in the ass if you loudly stated something that does not correspond to reality:-)
If you have a way of getting non-accidental BP from accidental, then you are either being slightly deceitful (e.g. looking into the future) or slightly deluded (underline as appropriate), there is no third.