How do you measure noise? - page 9

 
Владимир:

It is a pity that the topic of noise has been sidetracked.

Noise can be detected in advance over a certain time range with a high probability. There is nothing unusual about it.

Observe, identify a pattern and process it into program code.

If it is not difficult, an example. You can put it in words.
 
forexman77:
If you don't mind, an example. You can put it in words.
Noise is not visually recognisable in perspective, but can only be identified by software or seen on historical data when it is no longer relevant.
 
the noise within a topic clearly corresponds to its name, can be measured by the number of "noises"
 
Maxim Dmitrievsky:
the noise within the topic clearly lives up to its name, can be measured by the number of "noisemakers"
Another one, having heard the noise in his head, appeared at the noise in the heads of the noisy ones.
 
Yuriy Asaulenko:

Promised to post a picture of the noise track over the indicator picture on the 2-page thread tonight.


Noise in points.

I must say the result is a bit unexpected for me. I had expected something different.

The noise does not exceed 30 points. It is quite possible to work with it.


Of course it cannot be used on PVs, but since 1H it may be tried. If it behaves the same on larger timeframes and if it is not looking into the future, at first sight one and a half to two spreads of expectation per trade can be suctioned out of it. It does not seem to be much, but 40-60 points of four-digit spread is very good.


 
Vladimir Suschenko:
Your smoothing is in fact a frequency filter. And what you call noise in this situation may simply be a high-frequency (relative to the smoothing period) component of the signal.
Well a mouving by definition is a filter, a filter by definition extracts some part of a signal from a signal-noise mixture, what and how much distortion it introduces is another question. I wrote in the beginning of my post that "at first approximation...". Since no one knows the true signal and we don't even know if it exists, we have to use what we have.
 
Vladimir Suschenko:
This is an oversimplified approach that inherently assumes high signal distortion.
A more rational approach is retrospective analysis, with a step-by-step determination of the signal components. Schematically, it looks like this:
- It is initially assumed that the price movement in the future is influenced by several factors, e.g. day of the week, time of day, market condition (up/down trend, flat), important economic financial news, etc. The complexity of the model will depend on the number of influencing factors taken into account.
- We look for the dependence of price movement on each of the factors, which can be reduced to the form "Price vector=F{Factor(n)}". Factors, on the price dependence of which is not observed, are considered insignificant and are not considered further.
- We sum up the obtained dependencies in the chart and overlay it on the real signal. The obtained difference will be "noise" in our case.
But in its essence such "noise" is also a part of the signal, simply because of presence of significant influence factors not considered by us, we will be able to determine but cannot predict neither the character of "noise" nor any of its characteristics.
So I don't see the point in measuring noise. But it is my personal opinion and my approach to this question.
All this is nice, but practically is not feasible - what you suggest in econometrics language sounds like - identifying the trend, seasonal, cyclical components and then the subsequent analysis of residuals, as a rule - using autoregressive methods. In forex, so far, few people have been able to do that.
 
Олег avtomat:

The question itself - how do you measure noise? -- is incorrect, illogical, wrong.

The first thing to understand is that the input is a "signal+noise" mixture.

Why is it incorrect, illogical, incorrect? If you managed to separate noise from "signal+noise" mixture, then you just count dispersion, if it's legitimate, and voila - noise is measured!

Oleg avtomat:

How do you separate the "signal" from the "signal+noise" mixture? When solving this problem, identifying the "noise" should not be too difficult.

This problem is solved by methods of adaptive control theory.



Well, it's clear that before noise is measured it must be extracted, but concerning the efficiency of the adaptive control theory methods in the issue of exchange noise analysis, I have my doubts.
 
sibirqk:
It's all pretty, but it's not practical - what you propose in econometrics language sounds like - isolating the trend, seasonal, cyclical component and then analyzing the residuals, usually by autoregressive methods. In forex, so far, few people have succeeded.
Well, you can safely spit in your face and scratch out the shameless eyes of those who promise easy earnings on Forex. And yes, very few people are making it....
 
sibirqk:

Why is it incorrect, illogical, wrong? If you manage to separate noise from the "signal+noise" mixture, then count the variance, if it is valid, and voila - noise is measured!

Well, it's clear that noise must be extracted before it can be measured, but I have my doubts about efficiency of the adaptive control theory methods in the analysis of exchange noise.
And what was it that turned everybody to measure noise? How on earth did you "manage to separate the noise from the "signal+noise" mixture if the task of determining the signal is, as you say, "practically impossible"?