Zigzag indicator and neural nets - page 6
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
Yurixx
Thank you, corrected. indeed a mistake. So as not to waste time, start with Stratanovich, as all the others (Kalman, Wiener, Bucy etc.) are special cases
Yes, Yurixx, I see it now. But here's how to practically compile those very statistics for Elliottians, I have no idea yet (given the perversions of Elliott itself within EWP, which I noted in my reply to eugenk). Roughly speaking, an (Elliottian) wave does not necessarily end at the ZZ swing break point of the same scale.
So it is not the same scale.
I would just find a classical figure 5-3 on the chart of a certain TF, select the ZigZag parameter so it would reproduce this figure, and then calculate, on the one hand, how many of these figures on a specified interval of history and, on the other hand, how many trends of the size not less than a specified one (for example 5 waves in total) happened during this period of time in total. It's not all statistics of course, but it's simple and accessible, and shows the total share of the 5-3 pattern in the sum of all market patterns in this piece of history.
This is just the simplest idea of calculation. A number of subtleties need to be added to it, but these are subtleties of a technical nature, everyone can solve them on their own.
Been doing a bit of research on Adverse tactics recently. Asked the multipoints a question http://protoforma.com/news/?p=166#comments. By analogy with their answer, one can suppose that it is not always possible to "picture" the alternation of Elliott waves on the same timeframe. On the other hand. If we proceed from relations of waves based on fibs, then, theoretically, it is possible, choosing a timeframe, to find where these relations (fibos) correspond to necessary values for corrections or extensions. And this will be the "true" timeframe, on which the current wave develops. I will mention such an idea a little in passing. On fibs. If the current price movement does not fit into any fibo levels relative to significant extrema, we should look for other significant extrema, perhaps at another timeframe where the movement will fit... And then one may argue (or not) about the significance of extrema found by the zigzag. The current extremum, where the last ray of the zigzag ends, is often of dubious value (more precisely, it can be valuable only to experts). Past extrema, on the other hand, are of predictive value.
The picture shows the last extension of 161.8%. And the Fibonacci level (50%) coincided with the level of the author of the largest branch of the parallel forum...
This is just the simplest idea of calculation. A number of subtleties need to be added to it, but these are subtleties of a technical nature, everyone can solve them themselves.
The market at the pre-Foreh stage (stocks and their indices) was indeed much simpler (and that's what EWP was designed for in the first place, as there was no Foreh). There were a lot of classic pieces (say a really classic 5-3-5-3-5 momentum - no overlaps between 4th and 1st, no failures of 5th, no diagonals in 1st and 5th). Now they also occur, of course, but the percentage of exceptions to them has grown a lot. So the classics no longer rule here as they used to, and the likely candidates for exceptions to even the rules (not just norms) are more and more common. This complicates the task of collecting statistics to such an extent that nobody has solved it so far, and no clear and convincing solution is in sight.
P.S. Here are more recent candidates: a clean five as wave B, a zigzag with the second wave in it bigger than the first, a triangle on the eur weeks in a place where it should not be in theory. These are of course only hypothetical markups, but the Russian EWP coryphaei(Akelo, Alexander FXO) are seriously talking about them.
Yurixx
Thank you, corrected. indeed a mistake. So as not to lose time start at once with Stratonovich, as all others (Kalman, Wiener, Büsi etc.) are special cases.
Thanks for the links. Stratonovitch is awesome! When I see such coryphaei, I am proud of the Soviet school of mathematics. It's amazing, the man saw the unity of physics, mathematics, computer science, cybernetics and statistics back in the 60s. It seems that the mathematical apparatus for creating a market theory is ready. What is missing is a very little - a person who is able to perceive the whole harmony of abstract constructions and physical understanding and to apply it to the confusion, chaos and bedlam of social interactions, economy and the market. :-)
Unfortunately, I did not find two books by Stratonovich on the Internet: "Principles of Adaptive Reception" and "Elements of Molecular Physics, Thermodynamics and Statistical Physics". And I did not find the article "On a priori - conditioned quasi-optimal filters" either. If it's available in electronic form, I'd appreciate it. Or a link.
If it were that simple and it would only come down to technical rather than ideological problems...
The market at the pre-Forex stage (stocks and their indices) was indeed much simpler (and that's what EWP was designed for in the first place, as there was no Forex). There were a lot of classic pieces (say, the really classic 5-3-5-3-5 momentum - no overlaps between 4th and 1st, no failures of 5th, no diagonals in 1st and 3rd). Now they also occur, of course, but the percentage of exceptions to them has grown a lot. So the classics no longer rule here as they used to, and the likely candidates for exceptions to even the rules (not just norms) are more and more common. This complicates the task of collecting statistics to such an extent that no one has solved it so far, and no clear and convincing solution is in sight.
Alex-Bugalter
You are a bit inconsiderate. If you're going to build a system whose main tool is the NS, read the theory of recognition. NS is just a tool and in order to use it correctly you need to have knowledge in this area.
I'm afraid it's you who misunderstands me. When it comes to zigzags, wave theory is automatically implied.
Using a pure zigzag, as a reference in a system, is not very acceptable, though it is beautiful. It is already a practice. It's a pity, but there's not a line about it in the books. It would have saved me a lot of time.
Allow me, based on my practical experience, to disagree with all those who consider the Sig-Sign to be of little practical use. I made my first Zig-Zag in Matlab about seven years ago, though I must admit that I only succeeded in predicting its missing ray to the current moment in time this year. Recently I have moved away from Zig-Zags and made a more informative indicator, more dynamically reflecting the trend, but its essence is close to Zig-Zag, so I present it as an example in the figure, to dispel skepticism about Zig-Zags and similar, and about their predictive power with .
On the picture the rays up to the last break point are plotted on history, from the last point to the current moment - forecast usingneural networks, the forecast is not of absolute values, but of the direction of the appropriate trend, the forecast is performed at the arrival of each new bar, neural networks are working in the mode of permanent re-learning.
Curves corresponding to M1 - red line, M5 - blue line, M15 - yellow line, M30 - pink line are plotted in one chart. In the upper left corner under the % sign the modelling and forecasting error of the corresponding curve is given, now it's equal to 0, though often different from 0, which allows to see in advance if the given forecast can be trusted.
The curves corresponding to M1 - red line, M5 - blue line, M15 - yellow line, M30 - pink line are plotted on the same graph. In the upper left corner under the % sign the modelling and forecasting error of the respective curve is given, now it is 0, though often different from 0, which allows you to see in advance if the given forecast can be trusted.
The curves corresponding to M1 - red line, M5 - blue line, M15 - yellow line, M30 - pink line are plotted on the same graph. In the upper left corner under the % sign the modelling and forecasting error of the corresponding curve is given, now it's 0, though often different from 0, which allows you to see in advance if the given forecast can be trusted.
I don't know MQL very well, so it is difficult for me to make a good Expert Advisor but I know how to set optimal work parameters for it. In addition it's impossible to use the Strategy Tester for my Expert Advisor system as it uses neuronets and an optimization unit with more than one hundred threshold factors for the indicator, the calculation takes about 40 seconds and it's impossible to speed it up.
I made the most primitive Expert Advisor, included an indicator that prepares information for using the neural network, it has no forecast but builds trend models without any adjustment or optimization, I didn't choose any operation modes, I show you what I got at once, the result is not very good, though if I improve the Expert Advisor, for example, prohibit opening another position in the same direction, in case of triggering of stop or trailing stop, I noticed it during visualization testing, the result will be much better.
Once again, this report does not refer to the expert system whose charts are shown in my previous post, but only to the performance of its input block, which can be tested with a strategy tester.
to Piligrimm
Zig-Zag is a good indicator that has an interesting property. It's just that everyone sees its use differently. Using it in its pure form on the NS input is useless IMHO. You confirm that too, that you have had to refine Zig-Zag.
I see its greatest usefulness when used as a breakdown of the quote stream into classes (which the NS has to recognise)
Piligrimm could you please clarify, if you don't mind.
.. .the direction of the corresponding trend... What is the output in your NS (what do you mean by trend ?).
...% modeling and forecasting error... What is your model and what is the time interval for which you are able to make a forecast.
It would be interesting to seethe historytest results over several years.