Building a trading system using digital low-pass filters - page 5

New comment
 
It looks like I just did the BGS a little wrong I did it right - the analyser hangs up. I'll think about why tomorrow. I don't think spectroanalysis is a questionable method.
 
Vinin:
Piligrimm:
At one time I had to work with random processes, and the task was to get at least an approximate prediction of time series with 90% of the random process component. To turn a random process into a quasi random one, I invented a simple method, I multiplied the random process by a deterministic process with close frequency-time characteristics, a sine wave in the simplest case, but better to use a more complex signal. As a result, the predictability of the process went up by orders of magnitude.

Can you tell us more about it? If you are willing to share, of course.
The essence of the idea is simple, when you create covariance of a random process and deterministic one, the resulting one inherits signs of both and becomes quasi random. And if we use several deterministic processes, different in their characteristics, for covariance, and create covariances with one random process, and then select the most informative features from the resulting group using genetic algorithms, and then make a forecast using the received signals, and as a result, subtract a deterministic process from the obtained forecast, then we will have a forecast of a random process in the residue, and the forecast accuracy will be much higher than in any attempts to make this forecast directly from the signals.
 
mql4-coding писал (а):
It looks like I just did the BGS a little wrong I did it right - the analyser hangs up. I'll think about why tomorrow. I don't think spectroanalysis is a questionable method.

As far as I understand, the analyser is from Finware ?
 
D500_Rised:
mql4-coding wrote (a):

Looks like I just did the BGS a bit wrong Did it right - the parser hangs. I'll think about why tomorrow. I don't think spectrum analysis is a questionable method.


As far as I understand the analyzer is from Finware ?


Yes
 
Piligrimm:
Vinin:
Piligrimm:
At one time I had to work with random processes, and the task was to get at least an approximate prediction of time series with 90% of the random process component. To turn a random process into a quasi random one, I invented a simple method, I multiplied the random process by a deterministic process with close frequency-time characteristics, a sine wave in the simplest case, but better to use a more complex signal. As a result the predictability of the process would go up by orders of magnitude.

Can you tell us more about it. If you want to share, of course.
The essence of the idea is simple, when you create covariance of a random process and a deterministic one, the resulting one inherits the features of both, and becomes quasi random. And if we use several deterministic processes, different in their characteristics, for covariance, and create covariances with one random process, and then select the most informative features from the resulting group using genetic algorithms, and then make a forecast using the received signals, and as a result, subtract a deterministic process from the obtained forecast, then we will have a forecast of a random process in the residue, and the forecast accuracy will be much higher than in any attempt to make this forecast directly from the signals.


Can you translate it into Russian, because you are inadequate
 
mql4-coding, you have such beautiful boxes on your website, do you get a box too when you buy it or just a picture of it?
 

mql4-coding

Just don't leave this thread, there are people here who can help and many of the forum members are as good as maths professors.

 
Piligrimm:
The essence of the idea is simple, by creating covariances between a random process and a deterministic one, the resulting one inherits characteristics of both and becomes quasi random. And if we use several deterministic processes different in their characteristics for covariance and create covariances with one random process, and then select the most informative features from the resulting group using genetic algorithms, and then make a forecast using the signals obtained, and as a result subtract a deterministic process from the forecast obtained, then we will have a forecast of a random process in a residue, and the forecast accuracy will be much higher than in any attempt to make this forecast directly from the signals.

Funny, but reminiscent of pulling yourself out of a swamp by a pigtail. I'm not being sarcastic - I just don't get it. Mr. Piligrimm, could you please explain in more detail: how to calculate the covariance of a random and deterministic series; by what principle (criteria) using the genetic algorithm you select the most informative characteristics (and characteristics of what); how you restore the initial series thus optimized and make a forecast. What is the basis for your claim of increased accuracy of such a prediction? Can you show these prediction results compared to direct "head-on" prediction? Thank you.
 

Kravchuk's system, in my opinion, is not remarkable in itself. It is built on the "standard" model of interpreting indicator signals.

Its peculiarity, which stimulated the development of the financial market analysis method in a way we already know, is the introduction of the nonstandard indicators (Numerical TFT).

But now it's clear that the author (and not only he) of this topic is looking for the more acceptable variant of the spectral analysis of time series. The topic is undoubtedly necessary and very interesting, but it requires sufficient skills in this area.

So we will have to wait for the learned men

 
Piligrimm:
The essence of the idea is simple, when creating the covariance of a random process ...
With a message like this, one can only see the simplest essence of the idea: To say a lot and not understand it, without saying anything at all.
123456789101112...32
New comment
Reason: