a trading strategy based on Elliott Wave Theory - page 42
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 agree! :o) Everything is already chewed up - now only those who can ;o) may swallow it.
We're only talking about methodology without going into details.
I can't answer more precisely, and I don't have much time to waste yet - I'm still trying to implement several approaches to strategy building.
Vladislav, thanks again!
I hope my approach will serve as a point of convergence between you and Yurixx.
Dear Yurixx!
You wrote:
Linear regression removes only the linear component of the original data, i.e., simplistically, reduces the overall RMS by the RMS of the linear regression.
If there was a non-linear component, the lack of time dependence of the RMS of the detrended series is not obvious.
As an illustration, we can look at GBPCHF W1 from 2002.03.31 to today.
At the final interval here the RMS seems to be decreasing.
Thanks.
Спасибо, Rosh, все понял. Как я понимаю, Вы экспериментируете на массиве в 1000 баров, а выборка для рассчета канала имеет фиксированную и, естественно, меньшую длину. Или Вы строите канал на всей 1000 ?
The quality of the sample will be essential here. By selecting a "off-trend" length you run the risk of capturing part of the trend that has gone away, or of under-sampling part of the current one. With all ensuing consequences.
Good luck and good luck with passing trends.
I cycle through channels from 45 (it is clear) to 1000 bars. (1000 bars on 15-minute charts is 10 daily candles, I think this is enough for 15-minute charts). On these 955 kanals I find RMS values and choose the channel where the RMS is smaller (at the moment), although I do not forget about the potentiality principle :) True, I haven't applied this method yet - one, and visually this method of selection doesn't always capture distant channels - that's two.
Perhaps, when I organize the top three channels, many questions will disappear as unimportant.
I encourage further cooperation/development without putting out the codes. The methodology is enough and much more interesting. As a last resort, we could have a one-to-one exchange.
Dear Yurixx!
You wrote:
Linear regression removes only the linear component of the raw data, i.e., simplistically, reduces the overall RMS by the RMS of the linear regression.
If a non-linear component was present, the lack of time dependence of the RMS of the detrended series is not obvious.
As an illustration, we can look at GBPCHF W1 from 2002.03.31 to today.
At the final interval here the RMS seems to be decreasing.
Thanks.
Wrong. Two random series that differ only by the trend linear component have the same variance. The figures in the Excel file confirm this, in fact I deliberately put a red arrow to accentuate .
Спасибо, Rosh, все понял. Как я понимаю, Вы экспериментируете на массиве в 1000 баров, а выборка для рассчета канала имеет фиксированную и, естественно, меньшую длину. Или Вы строите канал на всей 1000 ?
Здесь существенным будет качество выборки. Выбирая "отфанарную" длину Вы рискуете захватить часть ушедшего тренда или недобрать часть текущего. Со всеми вытекающими последствиями.
Удачи и попутных трендов.
I cycle through channels from 45 (that's understandable) to 1000 bars. (1,000 bars on 15-minutes is 10 daily candles, I think that's enough for 15
minutes.) On these 955 kanals I find RMS values and choose the channel where the RMS is smaller (at the moment), although I do not forget about the potentiality principle :) True, I haven't applied this method yet - one, and visually this method of selection doesn't always capture distant channels - that's two.
Perhaps, when I organize the top three channels, many questions will disappear as unimportant.
I encourage further cooperation/development without putting out the codes. Methodology is enough and much more interesting. As a last resort, we could have a one-to-one exchange.
I've already written briefly: build swings on extrema starting from about 180 days ago. There is no need to go further - the result will be the same. All the bars that make up the trend should go from extremum to extremum (extrema should be included in the reversal zones ;) ) - then it is a matter of technique - you identify the last active channel and lay it out. From there the degree of nesting or detailing.
Choose from the resulting subset.
Good luck and good trends.
The values at the beginning and at the end of the red arrow in your file are different, and usually the greater the difference, the stronger the slope of the linear regression.
However, the difference is not equal to the numerical RMS of the linear regression, and speaking about this decrease in Yurixx I meant, figuratively, the cause of its occurrence, not the method of estimation.
I did not immediately see that the difference turns out to be insignificantly small, and indeed they should not be different, since the linear component is removed when calculating the RMS by taking into account the expectation of error.
But in my post I did not mean RMS of LR errors, but RMS of LR line. And also the RMS of the raw data, not the RMS of the differences between successive values.
I apologise for being unclear again.
Choose from the resulting subset.
Good luck and good trends.
We forgot about the elephant, thank you. Sometimes there is such psychological blindness, when you can't see the forest for the trees. I have already made a script that puts channel bounds (first and last bar), the only thing left to do is to teach channels to rebuild according to movement of these vertical lines, but it's not difficult.
Something like this zigzag ?
I agree. Although it would be better to do without code fragments.
For example the picture Vladislav told me more than any fragment.
And in general I believe that the question of distribution of the system, as well as its commercial (not for personal use) is the sole prerogative of the author of the method.
Rosh, in principle the derivation of the equations themselves is obvious. Everything is clear with that. But I understand that you are using averages for x and y. That is, you are simply solving one equation by clear methods of linear algebra. But what I don't understand is the following. Is it really possible to simply substitute sample averages into these formulas and get exactly what we need? Could you give a proof of this? Does the ANG3110 indicator work according to this principle? I think
it would be more logical to solve N such systems for N bars and from the sample of obtained arrays a,b,c determine the expectation of each parameter and use it as a parameter for the approximating parabola. Or am I mistaken?
There was such a thought, in principle I even thought of averaging the linear regression coefficients by this methodology, but so far no luck. Is it worth digging in this direction or not?