Building a trading system using digital low-pass filters - page 20
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Oh, shit, I lost all my links, anyway - there was a forum, quite a long one, where the guys got serious about two things
On Alpari or Viac, a thread with a title something like "filtering bourgeois bazaars" - that's probably what it's about.
Oh man, I lost all my links, anyway - there was a forum, quite a long one, where guys took up two things seriously
On alpari or viac, a thread with a title something like "filtering bourgeois bazaars" - that's probably what it's about.
to Neutron
What if the input characteristics of your reference filter are not chosen correctly or in the most sub-optimal way?
Yes, I tried to change the averaging window in a very wide range - it did not affect the result at all, or it did, but very insignificantly. We count CO of the studied filters relative to it and then normalize to the CO of BP relative to it.
Addition to what has already been written, to grasn's remark.
I got lost!
On a choice of a window at LPF depends magnitude and position of maxima at researched rows. Does not depend or weakly depends on their relative position.
It turns out that at first you need to pick such ELF that clearly shows on BP the points of interest, and then select the desired window of the conventional LPF. This will be the optimal choice in the sense of maximum probability of revealing the required patterns.
Sorry for my "technical illiteracy" - what did you just write about?
LPF has quite a few parameters, of course they are fully defined by the specification but still, there are quite a few: sampling step, bandpass/suppression limit frequencies, bandpass/suppression nonuniformity factor, etc. Which window are you talking about? If you're talking about representing a filter specification as a single input parameter, then...I hope you don't use such a filter in real trade?
Maybe the filter was not rebuilt properly? Butterword has no input characteristics, like a window, but similar to a "window" - calculated coefficients, which are fully determined by the specification. Where do you have the specification??? Chances are you've just fixed some characteristics and are now making discoveries, congratulations.
And it's the relative position of the extremes that is independent or weakly dependent? And it's not even dependent in a poorly designed filter, or a well made one, but not for "that" signal? Cool, give me two such filters...
Well, yes! Butterworth filter has no averaging window - it is recursive, and there are many knobs - parameters that can be changed, thus determining the slope, unevenness of frequency response in the passband, the passband itself... But if you want, you can simplify all this variety to one knob and thus get the main idea.
And about position of extrema depends on WIDTH of LPF bandwidth - I was corrected!
The approach itself seems promising. A man comes and shouts, "Here, I've invented a super-duper cool VFD! You give it to him, let's see how it smoothes. And we compare it with, for example, the moving average. If it produces the best smoothing, respect to the author!
The kicker is that we were able to identify a single generalized parameter for all LPFs allowing for their objective comparison, and this parameter is the deviation of the smoothed BP from the perfectly smoothed LPF without LPF with straight frequency response in the pass band, etc. Of course there is some arbitrariness, but nothing better can be seen.
here's http://www.bcs .ru/school/prof/mts/2003/Gorchakov.zip, by the way, I recommend it to everyone, especially those who like stationary. There is also some information about filters, by the way.
The nice thing is that it's about where my thoughts intersect. :)
here you go, by the way http://www.bcs.ru/school/prof/mts/2003/Gorchakov.zip
It turns out that the optimal statistics for building our trading systems in this situation are linear moving averages with variable windows. I.e. moving averages, which should be taken in such a way, that they mainly fall on the sections of one trend. And the moving averages are not exponential or anything else, they are 2 moving averages: a simple moving average and a moving average with factor i, which is the sum of i by Xi. And for volatility you have to look at the sum of the squares of the sequence of these random variables.
Looks like the author has come close to discovering LRMA :)
The LRMA is designed so that the sum of the squares of its deviations from the price is minimum. But another target function (TF) can also be minimised - the sum of error moduli, for example. This TF, imho, is more natural for forex than the sum of squares of errors. It is problematic to calculate it analytically, but you can try to approximate it.
This is an analysis by difference substitution. I'm not discarding the idea though :-) You can increase the number of bets and their duration by adding filters on small timeframes
here, by the way http://www.bcs.ru/school/prof/mts/2003/Gorchakov.zip
It turns out that the optimal statistics for building our trading systems in this situation are linear moving averages with variable windows. I.e. moving averages, which should be taken in such a way, that they mainly fall on one trend segments. And the moving averages are not exponential or anything else, they are 2 moving averages: a simple moving average and a moving average with factor i, which is the sum of i by Xi. And for volatility you have to look at the sum of the squares of the sequence of these random variables.
Looks like the author has come close to discovering LRMA :)
I don't think that man (one of very few, by the way) who was able to adapt solution of stochastic process disintegration problem (just our case, data are conditionally stationary) to the market is not aware of LRMA. :)