Predicting the future with Fourier transforms - page 52

 
alsu:
Where did you read about edge effects, can you give me a link?
I found it. Vorobyev V.I., Gribunin V.G. - Theory and practice of wavelet transform.
From p. 90 it's about lifting scheme, on p. 95 it's about "boundary problem for signals of finite length".
 
AlexeyFX:

Fourier is no good for that either, for the same reason.


Why is it not good? You said it yourself, and I agree with it, that if you understand the nature of overdrawing, then the overdrawing indices are "not harmful".

Fourier has the same redrawing, only it is probably caused by jumping sampling frequency at different levels. In fact, Fourier decomposes the series into the sum of multiples of harmonics, but the segments of semi-periodic axes of harmonics fluctuate relative to each other, so the harmonics do not add to each other like a matryoshka, but jump, the spectrum is smeared, the segments of spectrum constantly overlap unevenly, climbing ends on adjacent. spectrum rattles. therefore within the decomposition mechanism we must align sampling rates.

 
Freud:


Why isn't it good? You said it yourself, and I agree with it, that if you understand the nature of overdrawing, then overdrawing indices "isn't harmful".

Fourier has the same redrawing, only it is probably caused by jumping sampling frequency at different levels. In fact, Fourier decomposes the series into the sum of multiples of harmonics, but the segments of semi-periodic axes of harmonics fluctuate relative to each other, so the harmonics do not add to each other like a matryoshka, but jump, the spectrum is smeared, the segments of spectrum constantly overlap unevenly, climbing ends on adjacent. spectrum rattles. therefore within the decomposition mechanism we must align sampling rates.

"nature of overdrawing" = the nature of the fluctuations of the series itself

practically 100%.

 
alsu:

The "nature of the overshoot" = the nature of the fluctuations of the series itself.

practically 100%.


There is no extrapolation for non-stationary processes in publicly available sources, but that does not mean that it does not exist in nature, I thought of this,

The problem is that the signal is not predictive, but rather a signal that corresponds to the "noise".

If we go in the opposite direction, originally using the available series to make a forecast for the present time instead of future prices, then mark points in the past that correspond to the desired forecast and extrapolate them, there will also be re-rating, but now we will analyze exactly the necessary process for profit, and for different periods, and then overlay the results of the desired forecast on each other and compare them to the truth.

 
What does non-stationarity have to do with it? Stationary series are just as good and are no easier to predict.
 
alsu:
What does non-stationarity have to do with it? Everything jumps no worse on stationary series and it's no easier to forecast them.


It means that the nonstationary series should be divided into a set of stationary segments (at different frequencies) or "stationary" combination of nonstationary segments, so it is probably more correct to say.

or more accurately, to analyse not a cloud of all forecasts in the history, but a cloud of variants that meet the desired boundaries.

Or, like this you can imagine.

 
Freud:


There is no extrapolation for non-stationary processes in publicly available sources, but that does not mean that it does not exist in nature,

The problem is that the signal is not predictive, but rather a signal that corresponds to the "noise".

If we go in the opposite direction, originally using the available series to make a forecast not for future prices but for the present time, then mark points in the past that correspond to the desired forecast, and extrapolate them, there will also be some overcorrection, but now we will analyze exactly the necessary process development for profit, and so for different periods, and then we overlay the results of the desired forecast and compare them to the truth.


It's all been stolen before us http://www.altertrader.com/publications03.html
 
Rorschach:

Everything has already been stolen before us http://www.altertrader.com/publications03.html

where does the sampling rate equalisation take place ?
 
Freud:
The frequencies have periodic components and can be extrapolated, but they do not have a constant component.
It is necessary to decompose the residual between the removed constant component and the difference between the filter product of the price series and the periodic frequency extrapolation of the residual filter without the constant component.
 
Freud:

periodic components can be seen

"oscillations" and "periodic" are two big differences. I can see oscillations, but I can't see that they have a period. That's why it's not interesting.
Reason: