Quantum mechanical methods - page 8

 
Dr.Fx:
Once again: Fourier in principle does not give frequencies "present in the series". It gives an approximation to a DESIGNED frequency grid. This is a typical demonstration of a fundamental principle of the measurement theory: as a result we observe not a property of the object, but a convolution of properties of the object and probe (an instrument, or, in this case, algorithm).
Gives an approximation to sine or sosine functions, theoretically the analogue of Fourier decomposition to other kinds of functions is possible. Spectrographs exist, they show frequencies that are there, functions are periodic, approximation by periodic functions, not non-linear and non-periolic. Writing from a tablet, I apologise for the errors in the text. Filters, even if lagged, are of little use without knowing what frequency they should be tuned to.

All imho.
 

Some time ago I read an article about adaptive sliders. I made two filters based on it in MQL4 (Kaufman's AMA has a similar algorithm).

ER-if there is a clear trend the parameter tends to number 1.

SC - the closer ER to 1, the less period of the indicator (i.e., the indicator value itself is a period)

Files:
ER.mq4  3 kb
SC.mq4  3 kb
 
forexman77:

Some time ago I read an article about adaptive sliders. I made two filters based on it in MQL4 (Kaufman's AMA has a similar algorithm).

ER-if there is a clear trend the parameter tends to number 1.

SC - the closer ER is to 1, the less period of the indicator.


The more "inventors" appear in discussion it's not bad at all. The subject of filters should be strengthened with mateaparatum, imho. The input data should be normalized in order to get frequencies, and the frequencies should be passed into filter parameters etc. :-)
 
Lo083:
Gives an approximation to sine or cosine functions, theoretically the analogue of Fourier decomposition to other types of functions is possible. Spectrographs exist, they show frequencies which are, functions are periodic, approximation by periodic functions, not non-linear and non-periolic.
Colleague, your illiteracy is amazing. At least read what you are told.

1. Gives an approximation to a sine or cosine function - not an approximation. It is a complete decomposition. Exactly the same as the original function decomposed by the DFT method.

2. It is theoretically possible to analogize the Fourier decomposition to other kinds of functions. If the basis is complete, of course it is possible. You can at least expand it into a Lejandre polynomial, who prevents you from doing so. But I would be wary of calling it a Fourier analog. There are some purely terminological nuances.

3. Spectrographs do exist; they show frequencies that are there. - They do not. Spectrum analysers show an approximation to THEIR frequencies, predetermined - read above. So they show what they show. It has very little to do with "what frequencies are in the signal" - to the extent of the (known) influence of the probe (algorithm) in the resulting convolution. And in this there is also, if you will, a ratio of uncertainties. You can't know the frequencies exactly from a finite sample. The product of frequency resolution by time resolution is always one. However, various non-classical spectral analysis methods can successfully break this limitation and are able (if certain signal conditions are met) to provide 1 Hz resolution on a 0.1 second sample of signal.
 
Lo083:

The fact that there are "inventors" in the discussion is not a bad thing. The subject of filters should be strengthened with a mapper, imho. The input data should be normalised to get frequencies, frequencies should be transferred to filter parameters, etc. :-)
The article was about correlation. The more pronounced the trend, the closer ER is to 1.
 
forexman77:
The article was about correlation. The more pronounced the trend, the closer the ER is to 1.
I don't quite understand what frequencies you are talking about. The purpose of the filter is to produce a smoothed signal, not to eat up any "frequencies".
 
forexman77:
The article was about correlation. The more pronounced the trend, the closer the ER is to 1.
Correlation between what and what do you propose to look at in the market?
 
Dr.Fx:
Correlation between what and what do you propose to watch in the market?

Or rather a linear regression. Draw a line from point A to B and see how the quotes deviated from the straight line. The more noise, the longer the period and vice versa.

This algorithm can be applied not only to the price, but also to other series, that's why I suggested it. What is not a filter?

Here is the article

 
forexman77:

Or rather a linear regression. Draw a line from point A to B and see how the quotes deviated from the straight line. The more noise, the longer the period and vice versa.

This algorithm can be applied not only to the price, but also to other series, that's why I suggested it. What is not a filter?

here is this article

I don't care if it is a regression. What is the input data? Leading question: does it bother you that the correlation between EURUSD and GBPUSD is one and the correlation between EURJPY and GBPJPY is another? So what does it take to know the correlation between EUR and GBP? :-)))
 
Dr.Fx:
You could do a regression. What is the raw data? A guiding question: does it bother you that the correlation between EURUSD and GBPUSD is one, and between EURJPY and GBPJPY is another? So what does it take to know the correlation between EUR and GBP? :-)))

Well, you are mathematicians and physicists, so figure out how to find out the correlation)

You can divide the ER of one by the other, where the value will be closer to 1, in those areas there is more correlation.