Author's dialogue. Alexander Smirnov. - page 27
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
... presence of fast or slow harmonics, etc.
If you don't mind, please, describe it in more detail, how you select it, using Fourier transform, Walsh, MME or another way. Interesuyut your estimate how many oscillations exist at the same time and how these oscillations were isolated. Thank you.
Fourier analysis, DCT, or a set of first derivatives of ma or some MACD. So far I find it easiest to work with derivatives of ma or T3.
Fourier is best used relative to Linear Regression. Depending on the scale, the 3 max harmonics are sorted out, tuned to resonance and usually their projections ahead are overwhelmingly triggered.
For myself I chose either T3 or LWMA, although in the process of further work on the analysis system it became less critical which algorithm to use - it depends on the market form, number of oscillations, ratio of forward and backward waves, the angle of trend slope or its almost absence, presence of fast or slow harmonics, etc.
... Fourier analysis, DCT, or a set of first derivatives of ma-shares or some MACD. So far it's easiest for me to work with derivatives of ma or T3. Fourier is better to use relative to Linear Regression. Depending on the scale, the 3 max harmonics are sorted out, tuned to resonance and as a rule their projections ahead are overwhelmingly triggered.
And how do you use derivative mashups, maybe lin regression is also an option here? How high have you reached purely in this direction? Or is it just a part of the system that also includes wave analysis in one form or another (at least in combination with a primitive analysis of the movement of local price extremums)?
Hello!
It is very interesting to know the opinion of the pack.
Posted an article on page 14 https://forum.mql4.com/ru/10446/page14 on Optimal Tracking Filters by John Ehlers . I am impressed by it and it seems to be similar to it.
Although the material is as old as the world, it has not lost its relevance IMHO. The authors of the article state:
OTF uses bar highs and lows besides the close price in its calculations. The part of the indicator formula, responsible for the adaptation, uses bar highs and lows in the calculation, which allows estimating the additional noise factor, which neither Kaufman's AMA nor VIDYA use in their calculations.
This has the advantage of using a single smoothing parameter. Kaufman's AMA requires the trader to make a decision regarding the choice of values for three different parameters. VIDYA requires a trader to make a decision concerning the values of two different parameters. OTF, in its turn, requires a trader to choose a single parameter - an averaging period (or a smoothing factor). It not only makes it easier to use, but also allows to grasp and understand its essence faster.
Isn't the Fourier symmetric (along the vertical axis) with respect to history, then trying to limit it to 3 harmonics... - what in practice is this % of "the vast majority of cases" and what is the delay/slippage?
And how do you use derivative mashups, maybe lin regression is also an option here? How high have you reached purely in this direction? Or is it just a part of the system that also includes wave analysis in one form or another (at least combined with a primitive analysis of the movement of local price extremums)?
The point is that the Fourier is symmetric, while the market is skewed most of the time, and then the left and right parts do not correlate well with the signal. And when the regression is built first, and harmonics and their sum are calculated from the data around or relative to it, their periods coincide more accurately with the real signal.
The top one shows a single of the sum of 3 harmonics. Each of them individually has its own manifestation. But you can calculate more harmonics 7-12 and sort them by amplitude, picking out the 3 maxima. Generally, if it is well tuned in resonance with minimum SPR, percentage of matching turning points is very high, I haven't calculated statistics, somewhere between 70-80% by eye. I have not got to automate it yet, I have no time to do everything at once.
Derivatives from ma are taken like this: suppose we have period - p, then let's input derivative coefficient, say, k=0.3 and shift psm = k*p; and subtract shifted value from current average value, dma[i] = ma[i] - ma[i+psm];
So far, the heights have been reached on the tester, and this is not an indicator. In 2 years I've got the figure 10000 out of the depot, which is many times greater than anything I've ever seen in others - so what of it - it's a fitting.
I'm working on an analysis system at the moment.
The point is that the Fourier is symmetric, while the market is skewed most of the time, and then the left and right parts do not correlate well with the signal. But when regression is built first, and harmonics and their sum are calculated from the data around or relative to it, then their periods more accurately coincide with the real signal.
Generally, if it is well tuned with minimum SPR, the percentage of coincidence of turning points is very high - I did not manage to crunch statistics, somewhere between 70-80% at a glance. I have not got to automate it yet, I do not have time to do everything at once.
So far, the heights have been reached on the tester, and this is not an indicator. Well, in 2 years I've got 10000 from depo, a figure many times greater than anything I've ever seen in others - so what of it - it's a fitting.
No, Korey, LWMA is listed as Linear Weighted MA.
In this indicator everything changes, it's interesting to play around, so I did it and put METODa=4)))