a trading strategy based on Elliott Wave Theory - page 254
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
Interesting, Sergei. And something I particularly liked. Exactly - about the "wave function". :-))
Mathematically, of course, there are a lot of incomprehensible things.
But as a matter of fact, I would like to look at this. The pictures you have cited refer in general to a trend.
1. It would be interesting to see how the forecast behaves in relation to the price in the area of trend reversal, but there is one on a reversal.
2. It would be interesting to compare the forecast using your system with the forecast using conventional linear regression. Judging by the pictures the LR also works well in these cases.
In the pictures the hour bars are plotted along the x-axis ?
One more thing. Apparently the decoding needs to be corrected: "red crosses" and "blue squares" replace with "blue crosses" and "red squares". :-)))
I obtained an interesting result, studying the comparative structure of real EURUSD ticks of 2006 and normally distributed random series (2200000 ticks), posted on this forum by Sergey.
In a nutshell, it turns out the following:
The movements of the random variable are almost 1.5 times greater than the movements of EURUSD. By movement I mean the absolute value of change of CB or EURUSD for segments of a zigzag (size on the y-axis), plotted on data of a corresponding series. This result is almost independent of the scale of the zigzag and takes place from the tick zigzag to the largest ones I have plotted.
The ratio of the lengths in ticks (i.e. size on the x-axis) of the zigzag segments for CB to the corresponding lengths for EURUSD, in contrast, depends on the scale of the zigzag. For a tick zigzag, this value is approximately 1.43 and falls quickly to 0.72-0.75 as the zigzag scale increases.
All this means that movements of the real EURUSD price have a clearly expressed return character and the value of this return is much larger than it can be assumed from Pastuhov's results. At the same time processes in the real market are developing much slower (except for the level of ticks) than for ST.
Maybe such results are obtained due to specific properties of the series of CBs generated by Sergei? Although, if memory serves me correctly, he did it trying to reproduce in the best way possible the distribution of first differences for EURUSD in CB. If I am mistaken, please correct me.
On the mathematical side of the matter, of course, there is a lot of incomprehension.
But in essence I would like to look at this. The pictures you have cited refer in general to a trend.
1. It would be interesting to see how the forecast behaves in relation to the price in the area of trend reversal, but there is one on a reversal.
.
You very rightly pointed out!!!!! And I was sure you would be the one to see it.
It's bad at the moment. This is the only weak point. The skylining coefficient is not yet able to anticipate, I'm working on that now. In terms of the model, the assumption that further price movement will be within the confidence interval (CI) is not respected. And the DI, by its very nature in the model (admittedly, it is not published in full), is in some ways a constraint on price development.
The criteria for choosing the number of N samples is let down. Now this is a real weakness and a particularly difficult point. This is also something I am working on.
2. It would be interesting to compare the forecast using your system with the forecast using the conventional linear regression. Judging by the pictures the LR also works well in these cases.
Works much better, completely honestly, whether with or without pictures, because it's all for myself, not for reporting, because it's my money that will be managed by this algorithm :o)))) !!! (H+L)/2 is rather accurately predicted, and therefore I can very accurately (within reason) estimate the trajectory itself, which can not be done using LR. Gathering of statistics will be done in 1-2 weeks, it will be seen.
In the pictures the x-axis represents hourly bars ?
Yes, note that the prediction is built "on small": only on (H+L)/2 and on VERY limited samples.
One more thing. Apparently the transcription needs to be corrected: "red crosses" and "blue squares" replace with "blue crosses" and "red squares". :-)))
Right!!! :о)))))))))))))))))))))))))))))))))))))))))))
Corrected for history.
PS:
Maybe such results are obtained due to specific properties of the series of CBs generated by Sergei ? Although, if memory serves me correctly, he did it trying to reproduce in the best way possible the distribution of first differences for EURUSD in CB. Correct me if I'm mistaken.
Note only that if Sergei created the series by summation, it is not random. However, I've already written about it.
I suppose it doesn't need to. Don't force your coefficient to do something it can't do in PRINCIPLE. You've been doing Hirst for nothing. Use it.
And the number of N bars in the sample is not such a significant goal. You won't find a pivot point prediction on that track. IMHO
I suppose it doesn't need to. Don't make your quotient do something it cannot in principle do. You've been doing Hirst for nothing. Use it.
And the number of N bars in the sample is not such a significant goal. You won't find a pivot point prediction on that track. IMHO
For Hearst the sample is very small, i.e. a priori I will find anything but the truth.
I demand only one thing from the number of counts, that future bars should lie within the channel borders, and this does not always happen. That was the idea - if it does, then the model is more or less working properly, and if it doesn't, then it lies. And for it to fit - we should step back a couple of bars (figuratively speaking).
The only thing I require from the number of samples is that future bars should lie within the channel boundaries, which doesn't always happen. That was the idea - if it does, the model is more or less working properly; if it doesn't, the model is lying.
Sergei, who prevents you from using the right sample for Hearst?
Who is obliging you to take the same sample for both Hearst and the predictive model?
And what is the value of this model if it only works on plots and you cannot determine the boundaries of these plots?
От количества отсчетов я требую только одно, что бы будущие бары легли в границы канала, что происходит не всегда. В этом и была вся задумка, если укладывается, то модель более менее работает нормально, а если нет, то врет.
Sergei, who prevents you from taking the sample for Hearst as you should?
Who is obliging you to take the same sample for both the Hearst and the predictive model?
And what is the value of this model if it only works on plots and you cannot define the boundaries of these plots?
It's tricky, or rather not so simple. Hearst's index reflects a phenomenon (at least Hearst discovered no index at all) that only makes sense for a specific sample and the estimate is valid only for this specific sample, no more and no less.
One also has to take into account "resolution", which is related to sample length, calculation method and other details. Exactly for these reasons I "see the forest" in a strategic forecast but don't "see the trees", or rather what's under my nose. :о)
But I've already fumbled for research directions...I'm sure it will all work out.
Addendum...
Thought I'd clarify my thought a bit. If we take samples of 100 and 500 from the current count, will the values of Hurst index coincide, for example, on the fifth count?
I hope it's of some use to someone...
you've been quiet... are you working?
I have never come up with anything better than implementing a simple tester procedure with a fixed take and stop in the indicator code in mql4...
However, now the thought of repeating that feat makes me shudder...
(there was a self-learning indicator. i can post the code if you are interested).
Have to test everything in my mind at the moment - using Tesla's method. :)
Thought I'd clarify my thought a bit. If we take samples of 100 and 500 from the current datum, will the values of the Hurst index coincide, for example, on the fifth datum?
Sergei, I think this is an incorrect formulation of the question.
One should suppose that the fifth check is counted from the current one and the current one is zero ?
Obviously, they will not coincide. And there can be no coincidence for any iterative procedure that uses a finite piece of history of different lengths. So what of it ? It's the convergence of the iterative process, not the coincidence, that matters. If it converges, then you only need to choose the length of history that ensures the accuracy you need. And you don't need high accuracy, since we are dealing with a stochastic process and the accuracy of Hearst's calculation has little influence on the accuracy of the prediction.
:-)))