Building a trading system using digital low-pass filters - page 24
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
I'm working on it now.
It is somehow possible to break-even by eye guided only by spectral analysis. So it is possible to automate this process.
Some people are able to trade at breakeven based on spectral analysis alone.
Some people are able to trade at breakeven without it.
Somehow it is possible to trade at breakeven by eye guided only by spectral analysis.
There is no such thing as stationarity. At least in the sense that they are trying to apply to it here.
mql4-coding 27.02.2008 18:29
Пока я перелетал из киева в торонто тема разрослась как грибы после дождя.... Сижу, читаю, вникаю. :-)
It seems to me that there is an attempt to solve an unsolvable problem, rather than solve the practical issue of trading.
And so one wants to earn, even if without higher mathematics... :)
There's no denying that the ATCF method has the right to live. Well, even if we can't analyze a spectrum with sufficient accuracy, but nobody denies that there are certain resonances....
So there are developed, and not bad, digital filter generators, I'm not talking about Finvar's one.
Why not go the other way round:
- take as a basis... at least, for example, the rules formulated by Kravchuk, even if separately, first one, then the next....
- we write a elementary code
- let's race through history, optimizing only two parameters: cutoff frequencies, i.e. we look for their acceptable combinations...
- further it is obvious....
The question here is practically one: at what time interval such optimization should be carried out, so that it would not pull at trimming and the result would not be lubricated, and at what periodicity to correct the obtained result.
What can you say, scientist forum users? .... But don't kick me at once, I understand, that's actually what this topic is about :)
...
The question is practically the same: on what time interval such optimization must be performed to avoid adjustment and to obtain non-lubricated results and also to correct the obtained results with what periodicity.
In TF theory there is a concept of optimal digital filter, and there is also a matched filter. So the first one (optimal) is the best, there is no better one, and it must match the spectrum of the input signal. What you're trying to ask (optimization, time interval) is an attempt to build a matched filter on history in hope that the spectrum will be the same (and a matched filter initially loses out to an optimal one by definition). And if we assume that price is a stationary process, then yes, this optimization attempt can lead to positive results. If the analyzed price series is non-stationary, all attempts of optimization on the history are doomed to failure.
The question here is practically the same: on what time interval such optimization should be done to avoid adjustment and to give a non-smudged result, and with what periodicity to correct the obtained result.
In TF theory there is a concept of optimal digital filter, and there is also a matched filter. So the first one (optimal) is the best, there is no better one, and it must match the spectrum of the input signal. What you're trying to ask (optimization, time interval) is an attempt to build a matched filter on history in hope that the spectrum will be the same (and a matched filter initially loses out to an optimal one by definition). And if we assume that price is a stationary process, then yes, this optimization attempt can lead to positive results. If the analyzed price series is non-stationary, all attempts of optimization on the history are doomed to failure.
Got it all... Got it a long time ago. And another thing I've understood is that this problem is so far.... "difficult to solve", to put it mildly.
Only I'm not a master of the mathematical apparatus, like you, that's why I'm trying to get closer to practice by the means available to me :)
And you won't deny, that if all these sats, fatls, etc. are put on a chart (even standard, very long calculated ones), then there is a very nice picture in practical sense, but it is calculated by history, isn't it?
rider
This thread seems to have started with the fact that if we use TFs like FATL, SATL, etc., (these are TFs whose parameters do not change with time, there is no adaptation, all coefficients are calculated during design) then a good TS is not possible. And if there was a stationarity it would be possible to find such TF parameters that would work always, but as the author of these filters said somewhere they don't work.
It is necessary to recalculate these filters all the time, someone spoke about their recalculation on the fly. This will improve the situation but not to the end.
Look, read this thread, it's one of those few threads on this forum that is worth reading, I think.
rider
This thread seems to have started with the fact that if you use TFs like FATL, SATL, etc., (these are TFs whose parameters do not change with time, there is no adaptation, all coefficients are calculated during design) then you do not get a good TS. And if there was a stationarity it would be possible to find such TF parameters which would work always, but as the author of these filters said somewhere they don't work.
It is necessary to recalculate these filters all the time, someone spoke about their recalculation on the fly. This will improve the situation but not to the end.
Look, read this thread, it's one of those few threads of the forum worth reading, imo.
I have read it, only we are talking about the same thing, but we approach the issue from different sides, you from the theoretical and academically correct, and I from the practical.... that is why I asked my question about "intervals". Ultimately, you will input your super-optimal and super-ideal filter with the same time series and you will have the same question: reliable interval, where resonance frequencies are acceptably stable and are not smeared on the spectrum.... In years, yes, you can in years - that's what I do in my files, because it's boring to put such a base into indicators manually, let the machine work :)... I can lay out the tool, if anybody needs it, it's not mine, but in free access, so I don't infringe upon the copyrights.
By the way, has anyone seen anything perfect in this world? :)
In order to apply spectral analysis, you must first get rid of the gaps in the price chart.
Something like this
Used to be
became
but it's not very clear by what criteria to do it ;)