Moving average from moving average - page 3

 

What an interesting topic)).

This example shows better why TF is better than MA

 
Mathemat:

Now try to explain to me what the real point of such a great bandpass filter is, as I'm not clueless about filters.

)))And the salt is there. Imagine how many times you need to run Ma 5 (for example) to get double or triple MA 10 for example, or rather to get as they say here "price lag" the same. And to take into account some other amplitude peculiarities. Accordingly, the weights will vary asKokain wrote.

They wrote somewhere that the problem is in false signals (kinks), so a smoother curve - does it not solve these problems? But the waving is debatable, though applicable.

In general, if we take the FIR, ma or better, the fan of all ma from the fan, will give us another "measurement" to analyse the original price series.

 
ZaPutina:

This example shows better why TFs are better than MAs

MA is a special case of TF.

In general (from my point of view) we can distinguish two types of digital filters in use:

"trend filters" - filters with the sum of coefficients equal to 1 (these are all types of MA) - follow the absolute price value and are used for smoothing (the result by definition reflects the process dynamics not in the present but in some recent past, and is used in algorithms of "assumption of future" only indirectly, in the calculation that we know some other characteristics of the process);

"Oscillators" - filters with the sum of coefficients equal to 0 - reflect the level of local fluctuations, and are also used in strategies indirectly. As the sum of coefficients equals 0, the normalization is performed by standard deviation reducing it to 1 (as an example). Subtracting the MA value from the current price, we also obtain the oscillator.

We can assume that the square and triangular kernels of the conventional MA are non-ideal, and write an algorithm for fitting the kernel to any required pattern with any oscillation period.

 
I understand that MA is a special case of cKFs. But TFs have to adjust to the spectrum as well, a kind of "shamanism". Mashes can also be shammed in terms of RMS, for what purpose is another question. To draw conclusions about the future one should correctly decompose the past first of all, TF does it better. Even a simple example of running a wrench several times from itself is worse than another cue, for example by amplitude, there are no negative coefficients in the impulse response of any wrench.
 
Valera's back! Where have you been?
 

Here are a few averaging runs of the same type and period, the more history the more the graph looks like a sine wave

And you can't tell that the price has been decomposed, if you reduce the number of runs, it looks more like reality

same piece

 
ZaPutina:

Here are a few runs of averaging prices of the same type and period, the more history the more the graph resembles a sine wave

And you can't tell the price was spread out, if you reduce the number of runs, it's more like reality

same piece

Well here it is ))))

It all makes sense now.

And the above spectra is a total out.

And Junko will tell you that the whole tsos is built on z-1 and operations over that))))

SENK S, COLLEAGUES!!!!

DSP can even afford to do an anti-quote if desired, annulling the current one to zero - isn't that right ??? )))))

 
_new-rena:

Well, here it is ))))

It all makes sense now.

And the above spectra is a total out.

And Junko will tell you that the whole tsos is built on z-1 and operations over it))))

SENK S, COLLEAGUES!!!!

The COC can even afford to make an anti-quote if they wish, annulling the current one to zero - can't it ??? )))))

I don't know who thisJunko is and what he'll tell you there. I haven't said everything yet.

PS. No offence, the feeling is that you're not getting anywhere and you're running around the branches telling your opponents with results that it's all nonsense), the next branch with spreads has also interrupted you).

 

You could do the runs differently, take filters with odd periods and shift them by (N-1)/2 during the runs,

 

Naturally history needs to be deep, but you can do the same quality enhancement without requiring an increase in history.

Right?)

There will be redrawing, but have you seen in accelerated mode how this redrawing jumps, it will be a set of filters (if we use the example of dummies) with overlap 1-3, 3-5, 5-7, etc.

i.e. mashups with odd periods from 3 to, for example, 1025. Then we prolong our filters at the ends assuming that the price at the ends has not changed and make this assumption after each new run.

We will get the same number of runs - 1025, or rather 512 - for each of masks from 3 to 1025.

The extrapolation will then have to be negotiated.

But, those who do the same through Fourier-using Fourier not for extrapolation, but for decomposition, probably do an analysis also in the imaginary part before proceeding further.

Later we will talk about phase and phase methods. There are many methods, just the laws of volatility and its connection on different timeframes, and periodicity, or rather cyclicity, are used.

For example, we can add to the decomposition mechanism described above an element of preliminary equalization of bars' sizes relative to something else, or use TF and make weights dynamic, well...

I wrote above that TF filters are better not only because smoothness can be set, but also because functions concerning volatility can be set in the coefficients of the filter itself.

To be continued...