FIR filter with minimum phase - page 8

New comment
 

A filter is one way of smoothing economic data. If you look at the literature, the two filters that have the most use are Hodrick-Prescott and Kalman. However, there are a great many filters, all perfectly designed and applied in practice. Why such limitation of application of filters? The answer lies in a completely different area and has nothing to do with filters, phases or anything else.

If a filter has been applied to input data, the initial quote consists of two components: the result of filtering and some difference between the initial signal and the filtered result. Since in trading (unlike in radio electronics) we are always interested in the forecast, it is quite natural to ask: can we extend the filtering result forward? The answer lies not in filtering result, but in filtering residual (noise). If noise is stationary (mo and dispersion are approximately constant), we can extend filtering result, while dispersion will be an error of this prediction. If the residual is not stationary (has variable mo, which can be removed, and worse has variable and often very intricate variance), then the prediction is not possible, because the variance is about the past and has nothing to do with the future.

Conclusion: all talk about filter phase is meaningless if the filtering results in a non-stationary residual.

 
In other words, I wanted to make a set of cues that could be used to cut the spectrum of history into equal parts with a depth of e.g. 1024 bars. But then we also need to extend the lines of these filters into the future. For this purpose we will need to extend weight functions by extending their edges in the form of a decaying oscillation. But such filters will redraw. The task is to continue filter's response to influence in such a way that there would be something like compensation, e.g. when repainted on a new bar, it would be necessary to make the next filter change its phase and repaint down, and so on, the set of filters would not sum up repainted but compensate each other, at desire to have minimal remainder in the end, that is, during build as if the response of the following filter from the previous one compensated repainted from it. Accordingly the length of impulse characteristics (initially a bell in the form of a parabola will be from a triangle pascal (for odd filters), further the cut ends of these parabolas we will continue building/choosing parameter of damping and depth of cic filters of different periods so, that the sum of modules was minimum.) will vary, that demands participation (at shift of filters) of IIR filters at such way of extrapolation. Or a wide set of such filters. I will try to think up an example in more detail and describe it later.
 
EconModel:

A filter is one way of smoothing economic data. If you look at the literature, the two filters that have the greatest use are Hodrick-Prescott and Kalman. However, there are a great many filters, all perfectly designed and applied in practice. Why such limitation of application of filters? The answer lies in a completely different area and has nothing to do with filters, phases or anything else.

If a filter has been applied to input data, the initial quote consists of two components: the result of filtering and some difference between the initial signal and the filtered result. Since in trading (unlike in radio electronics) we are always interested in the forecast, it is quite natural to ask: can we extend the filtering result forward? The answer lies not in filtering result, but in filtering residual (noise). If noise is stationary (mo and dispersion are approximately constant), we can extend filtering result, while dispersion will be an error of this prediction. If the residual is not stationary (has variable mo, which can be removed, and worse has variable and often very intricate variance), then the prediction is not possible, because the available variance relates to the past and has nothing to do with the future.

Conclusion: All talk about filter phase is meaningless if the filtering results in a non-stationary residual.

All true, except for the noise! That's not noise. Let it just be the residual. The residual is non-stationary. But it may be insignificant for the prediction. Inertia exists for the entire spectrum.

============

Valera (Nik1972), stop cluttering up the thread!

 
Zhunko:

The residual is unsteady. But it may not be essential for a prediction.


Always? Or will it become significant when unplugged?

The residual should always be analysed and modelled if necessary. an unsteady residual should not be left. It seems to me.

 
EconModel:


Always? Or will it become significant with an internet outage?

The residual should always be analysed and modelled if necessary. you cannot leave an unsteady residual. It seems to me.

The prediction should be made on every bar. The residual will change, but not much.

How long can you extrapolate the MA or MACD line into the future without exceeding the set error?

You do not have to answer. You set this error and the range of the forecast. The TS works on these data.

What difference does it make whether the Internet is down or not? Put technical stops, follow the MM, do not risk too much.

 
Zhunko: Valera (Nik1972), stop cluttering up the thread!
Vadim, it's not obvious yet. But the style is similar.
 
Mathemat:
Vadim, it's not obvious yet. But the style is similar.

That's the one!

1. Style.

2. Mistakes.

3. Most importantly, the dynamics of learning. He reads something new somewhere and starts talking to himself with a smart-ass look.

 
The forecast is not constant in depth, and all known extrapolation methods are based on the static depth in extrapolation, and this parameter also floats, due to the floating spectrum, by constructing an accelerated change ("floatability", redrawability, or whatever) of the spectrum, you can forecast its state into the future.
 
Nik1972:
The forecast is not constant in depth, and all known extrapolation methods are based on static depth during extrapolation, and this parameter also floats, due to the floating spectrum, by constructing an acceleration of change ("buoyancy", redrawability, or whatever you want) of the spectrum, you can predict its state in the future.

So build it, what's the problem?
 
Peter_Zabriski:

Oh, Vadim! Almost philosophical. I almost agree. Keep going, please.

===
Don't forget Nyquist. You'll offend the old man...


You .....Nyquist... Screw you and you are asking for advice. Kotelnikov and his theorem is the MOST important + common sense.

Z.I. That's how I got banned till 2022, banned ))). I'm an old man, I'm offended. Can't get past it. The whole world has long recognised that the theorem is far more important than any frequency.

123456789101112131415...20
New comment