a trading strategy based on Elliott Wave Theory - page 250
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
Come on, it was a great joke to laugh out loud at,
patting yourself on the knees and pointing your finger. :)
Alexei, good day!
Alexei, let's flub a little more on-topic, shall we? :))
Please help me insert a picture, the explainer ("How to insert pictures in this forum (explainer)") doesn't quite explain something :)
How are you doing with your demo account? What is the balance already?
Regarding the past description of the chart, I meant:Euro, 1 hour
Grash If you use iterations in your algorithm you can try to use genetic algorithm. You may be able to speed up the calculations. Question:( If it's not a secret, of course.) I suspect that you are using some tricky wavelet transform method of your own. Am I right?
Besides, calculation of forecast takes 3-7 hours, depending on specific structure of the row itself... And this is quite a lot...
Grash If you use iterations in your algorithm you can try to use genetic algorithm. You may be able to speed up the calculations. Question:( If it's not a secret of course.) I suspect you're using some tricky wavelet transform method of your own. Am I right?
You're right, I do use wavelets, but all within theory, no amateurishness.
I'm working on a genetic algorithm at the moment.
EVA can be thought of as a set of chaotic dynamics action patterns.
Moreover, the set is far from perfect. there is a more advanced (and strictly justified) system of analysis - Tactica Adversa.
By the way, I'm not happy with the stochastic approach, in principle.
(i.e. how can it be that i don't know if i'm buying high or cheap...? it doesn't make sense; there's absolutely always a "fair price"... "s.c." per city, per country, per world, with/without taking certain parameters into account, but it _is_! anyway!)
Of the ideas of determinism, I understand that nonlinear dynamics is the most developed and fits our case...
Attention, question!
are there more efficient approaches? (networks/game theory/vibrations etc.)
Maths, unfortunately, I study very recently, so I don't know what else can be applied...
P.S. by the way, an amusing document on probability shift in chaos I got off the net... I'm putting it out there...
http://tovaroved.lv/nonlin/p7-14.pdf
P.P.S. If anything, the terminology is described in the textbook "Dynamic Chaos" by Kuznezov
My calculated pivot zone is represented as a rectangle and is relatively long in time. So, I had a need for mini forecasts, of course, because of "natural greed" and I started doing it "at my leisure".
According to my idea the mini forecast starts working as soon as it enters such a turning area. The only objective is to maximally define the local extremum for optimal trade knowing the forecasted direction of price movement and boundary conditions: the turning point area and the current price position in it. The algorithm for identification of a local extremum in the turning square based on the mini forecast is unlikely to be of interest for anybody, except for me. But the mini forecast itself is a very interesting task.
The matter is worsened by the fact that I cannot use well-tested method based also on Hurst's index - the sample size is too small and it's paradoxical - small value of forecast.
I've decided to investigate different variants one by one and here I'm sharing the results of my first try. So, I don't use autocorrelation and other tricks yet. For starters it's simple: the process is represented as a superposition of several pre-selected functions:
I have defined the following as such functions:
(1) Line of mathematical expectation (horizontal straight line)
(2) Linear regression
(3) Parabolic regression
(4) Harmonic
(5) and some more...
Since the forecast should be done for a small number of bars ahead, I assume that the basic patterns of each function will approximately hold. For example, the period found for the harmonic will still be preserved for a certain minimum number of bars. And the superposition of these functions should show "about - right" values. Yes, I know this will definitely never happen, but I definitely don't need it, and it might work.
The basic algorithm
(1) Find the optimal coefficients for each function by least squares method, of course, except for the mathematical expectation line :o)
(2) Find coefficients for a linear superposition of previously defined functions using the method of least squares
Remind me, the point of such prediction is not to play in a channel, but to find a local extremum. There is only one input parameter, it's the sample size for the prediction. I suppose the maximal forecast value should not be more than 1/3 of a sample. I use (H+L)/2 as input for the calculation.
Total number of samples in the test sample..............................................................50139
Number of current bar taken at random..............................................................................25000
Number of bars in the sample...............................................................................................18
So, I selected the current bar at random and this is what I got:
Black stepped lines - High and Low correspondingly, red solid line - calculated forecast function, red dashed lines - standard deviation from the forecast function, blue dashed line - mathematical data
The fact that we got a curve resembling a parabola is not surprising. ANC calculates coefficients in such a way that the function which is most similar to the data source "wins".
The results seem to be encouraging, but it is likely that the prediction must lie somewhere. Keeping the input parameter (number of bars in the sample), we shift forward to 25010 bars, assuming that it has already been formed. Immediately we see that the forecast lies:
It lies a lot. But! After I've done a dozen of experiments manually and wrote a small test for the whole sample it was clear to me that I could always find such an N from the current count that the forecast according to this scheme would show good results. The test was very simple: for each counting with the step +1, we increased the sample for which we had made the forecast and then checked how many future bars fell within the RMS limits. This test confirmed what has often been discussed on this forum. I did not find any sample for which it was impossible to find a sample and perform the correct forecast. For the sample 25010 there were as many as two such values N (adjusted, of course):
14
71
Now I'm thinking hard about the criterion for the number of samples. By the way, one such criterion can be seen "with the naked eye", if you look carefully at graphs. I am working on it now. But it is not enough one, I need to come up with a couple more.
Is anyone interested in this, or everyone keeps reading Pastukhov? :o)))
to Neutron
Sergey, where have you disappeared to? I'm tempted to write "don't sleep!!!". :о)))
It is interesting to me, though not actually for the price, but for the indicator.
And it seems that even the few people interested are disappointed in Pastukhov. And for nothing!
In the end, the discussion broke off at the most interesting point: how to make a working strategy out of purely mathematical
results to make a real working strategy. Never mind, who needs it?
You can be sure that EVERYONE is interested! Both Pastukhov and your research.
And why don't you use as red dotted lines not RMS, but limits of confidence intervals, constructed on the basis of 3 sigmas, or calculated by Student's for example for 99% confidence interval? Or do you have any special purposes for the chosen boundary construction? Just out of curiosity.