a trading strategy based on Elliott Wave Theory - page 138
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
"The purpose of the visit? - Lunch" (c) "Men in Black"
Ok.
:-)))
If a researcher sets up an experiment with a rat running through a maze, and if the rat runs the wrong way, the researcher is more inclined to believe that the rat is sick, rather than that the experience disproves his theory.
Является ли такой подход к автоматическому трейдингу перспективным и реализуемым ?
Это совершенно другой подход к торговле, который не может быть описан в рамках математической статистики.
Не согласен. Для каждого образа, на основе которого оценивается рынок, необходимо сделать статистический анализ поведения рынка при появлении этого образа на истории, построить распределение, определить его параметры. Далее можно опять-таки статистически определить оптимальные параметры образа. Имея эти данные в базе данных образов можно с известной точностью определять вероятность ошибки/успеха при принятии решений.
I think estimating statistical characteristics of patterns is rather questionable in practical terms, although information about attempts to do it can be constantly found on the Internet. For example, the latest version is presented here : "MQL4: The Self-Learning Expert Advisor" This method refers to neuronets. But for some reason I've never found any quantitative characteristics of such systems. Maybe I just watched badly?
And I think the problem here is that you simply may not have enough history to estimate parameters of individual patterns if you select patterns for estimation based on rather strict parameters. Or if you are going to statistically estimate ALL combinations of bars that can be conditionally attributed to the concept of a pattern, then I think you will either have a huge amount of all possible pattern modifications differing from each other by a very small statistically insignificant value, or the statistical estimates of patterns will be very smeared (a large variance of estimates). And in the future it will be very difficult for an automaton in real trading to understand which of the existing pattern modifications in the base that appeared on the latest bars now belongs to. For a pattern to be statistically significant, I think we need history on our working currency pair, which we do not have, and all this with guarantees that the market will play by the same rules in the sample that was not trained (in the future). In this respect a simple linear regression channel is statistically much better assured. Look at what is more reliable? Market condition information ALWAYS calculated with standard simple algorithm on the last 300 bars during history, or information that these bars obtained recently are similar (well correlated) to the averaged value of 100 instances of head and shoulders pattern, present in the history for the last 5 years? In my opinion regression is more reliable as it is a well studied and worked out mathematical technique compared to pattern recognition where there are too many dependencies on various other factors.
However I think the pattern recognition task can be reduced to a simpler trendline task (sloping resistance/support lines drawn along extrema). That is many classical patterns can be replaced with a set of trend lines breaking through which will mean working out of the pattern. But it is not that easy here as well. For example, in this file we can see the dynamics of a convergent triangle https://c.mql5.com/mql4/forum/2006/08/triangle.zip.
You can see the exit from the converging triangle on August 8. But according to the classical description of this triangle, the breakout should have been upwards only. But in practice the price went up and down, i.e. both bulls and bears got their money. This example immediately negates the meaning of the pattern "Converging Triangle" as such.
The trend lines in the given charts are drawn without taking into consideration the last 2 bars. That is why when a trend line is broken it is clearly seen which trend line was broken.
A more complete version of the trend lines dynamics for the last month can be found here "MQL4: A picture for the metaquotes forum" solandr 31.08.2006 08:02 (A multivolume RAR archive. There are 16 parts in total. After you have downloaded all parts, change the zip extension to rar and unpack it in WinRAR3.50. It's very useful for beginners in trading to watch this cartoon, for example ACDSee, to understand how the market trend may change over time and what can be done to minimize your risk.
In my opinion, working with trend lines is much easier than working with multi-parameter patterns that need to be captured on historical data and then gathered statistics. Trend lines are much easier! I even experimented with them in my Expert Advisor. I used them to replace a confirming oscillator and even the Hurst indicator! And in general, the obtained result was very meaningful, clearly different from a completely random one. For the time being, I decided to postpone using trend lines in my Expert Advisor for some time, since according to my observations, calculation of the Hearst indicator provides approximately the same information as trend lines do, but using a more formalized calculation algorithm that is more efficient in terms of creation and practical use of MTS.
Neural networks are most likely to be useful in the areas where you can calculate in advance all possible combinations that can happen in the future and just find a confirmation for one or another variant based on those combinations in the noise. For example, knowing beforehand (having recorded preliminary on a testing area, or having calculated all possible variants of signals on the basis of mathematical model adequate to situation) all possible variants of signals at approach of one object to another, in the future (at real use of the object trained this way) it is possible to find the closest from available variants of signals and to make the corresponding decision on further actions of the object where the system trained on neuronets is established. But all this works within the limits of what situations available in the database will happen and will evolve according to the once recorded algorithm. I'm afraid forex is more varied in this respect :o(
Yurixx, I think you were wrong to disagree with that.
Why not? I mean why do you think so?
Apparently, in summer science was simply on holiday ;o). And now an academic year has begun and there are some more appropriate status of a scientific forum:
http://www.lib.mexmat.ru/forum/viewtopic.php?t=3254&sid=a6468f00350c4e81f1b818be89ec1869
Thanks for the link, I will read it tomorrow.