a trading strategy based on Elliott Wave Theory - page 132
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
I would prioritise differently - the correctness of entries should be even around 50%, but stops and profits should give an advantage. In other words, we enter where we can take either a small stop or a large profit .
In the limit of small steepness of channels this would be the case. In general the picture is more complicated, because it overlaps with Yurixx' s answer, there will be some additional reasoning below.
I asked the question about your entry level because, to understand your approach to evaluation, I needed to see the ratio between SL and TP. Now I understand that it is 1:4.
I use current RMS levels, not fixed at the moment of entry, i.e. this ratio is true only at the moment of entry. Further, on a trend entry, SL starts to tighten and TP starts to move away. And vice versa, respectively, when entering against the trend.
Generally I imagine the options are:
1. Equilibrium valuation. SL = TP. I like this option because it is simple and gives an objective assessment of the "correctness" of the entry. That is, it gives an estimate of the system's increase in probability of winning.
2. Nonequilibrium estimation SL < TP. This variant allows you to estimate how close to the reversal point the system enters (for counter-trend entry) or how far it enters from the end of the trend (for trend entry).
3. Complex estimations. There are many of them, of course. And each of them can evaluate the specific property of the entries the system provides. Let me give just one example, which I also used. SL is not given, the only parameter is TR. For each entry the maximum drawdown that was reached before the entry reached TP is estimated. By varying the TP we obtain a series that can be statistically analyzed. This is just an example that has its disadvantages. In particular, ТР may never be reached at all. Therefore, the application of each such estimation variant requires its own refinement.
In general, when estimating the system as a whole, we rely on two values: the amount of positive trades for each negative one and the ratio of the average profit for profitable trades to the average loss for unprofitable ones. All these values are obtained as a complex when testing the system as a whole. Therefore, they are not independent in the sense that we cannot say why these results appear. Whether it's because the inputs are bad, or because the outputs are bad, or because the SLs and TRs are wrong, etc. So it would, of course, be great to standardise the methodology for assessing inputs and outputs (and they are related). Then it would be possible to build a methodology to independently evaluate the two main characteristics of the system. This would immediately show where the strengths of the system lie and what still needs to be improved.
I had a similar idea except I consider entering and following orders as degrees of freedom (which includes exiting). If they can be optimized separately, they will first of all result in reducing the amount of work (roughly speaking, the sum instead of the product of amounts). Indeed, in the standard definitions they are not independent, hence the need to reformulate them, although this may make the concepts less obvious and more difficult to apply. That is, a kind of orthogonalization would be needed. In this sense your third option, by the way, seems quite interesting, at least as an introduction to reflection.
I cannot say exactly which algorithm would be correct to solve this problem head-on (although I got interested myself), but here is an approximate algorithm.
It is clear that SL and TP levels may change. It is also clear, it may be a result of changing of calculation parameters and MM, i.e. trailing, etc. However, in order to standardise and study the efficiency of the inputs (and outputs) all these things must be eliminated. If good results are obtained on standard and fixed conditions, then a sensible strategy and MM can further improve them. If good results cannot be obtained by themselves, then MM will only smear this warning picture.
I've also tested the differences in the result with different sample lengths. If interesting, the results were presented here "trading strategy based on Elliot's Wave Theory" solandr 23.06.06 10:36
The balance graphs at 300 and 1000 bars have quite high correlation. Frankly speaking, apart from linear regression I also use parabolic regression (I wrote about it a couple of times already), and then I average data obtained using both of them based on information from several sources that this method allows to get closer to "true" parameters of channels (borders of real existing channels (and not those we choose in calculation).
I would also like to remind the esteemed branch members that strategy is an aggregate of several components (i.e. it is impossible to go far on just one linear regression channel, even though the channels will absolutely coincide with those defined by Vladislav on the basis of minimum potential energy!) I do not know whether you use Murray's levels in your EA or not, but they play an important role, as well as the Hearst Ratio and money management (the latter, as Yurixx rightly noted, is relevant to a greater extent only for maximizing the profit, which the other components provide). First, I can recommend you to enter the market by the most stringent conditions, softening them later (to increase the number of deals and consequently the total profit) as the position management algorithm is practiced. In other words, strategy success is half a result of the methodology described in this thread, and half a result of a successful (or more precisely, rational) position management algorithm. And the question of whether the position management algorithm is successful or not can be answered independently through the strategy tester. And exactly the second half is what everyone can (or can't) find by oneself and what Vladislav refused to present to the public from the beginning and it is not only about the potential energy, around which there are very interesting debates in this thread.
Then this question: do you do test for statistical significance of linear and parabolic approximation parameters?
That is, for some sample we have a good approximation for linear regression Y=A*X+B and for Y=A1*X^2+B1*X+C. We need to check whether these approximations are approximations of the same order. If they are, then parabola can be easily rejected as artificiality, if they are not, then we have two different approximations of one and the same time series, and this can serve as criterion of linear regression channel break.
No I don't, although of course this idea should be checked too. The parabola and the straight line approximate the price series exactly in the limits of their possibilities. But the parabola has more approximation possibilities because it is more "powerful" (of order 2, while the straight line is of order 1). That is, depending on the selection the parabola can transform into a straight line, but the straight line will certainly never transform into a parabola. Although an order greater than 2 cannot be used for approximation also because there is a stable opinion that higher approximation orders approximate not a trend but already random components of the trend, thanks to which Vladislav can therefore claim that the type of trajectory is not important and two curves lying in a given range are equivalent in terms of approximation.
Regarding the criteria of breaking of the linear regression channel I also have so far purely visual observations that a parabola top is formed before the break of the linear regression channel. That is the trend reversal point can often (but not always) be approximated by a parabola having the peak in the trend reversal zone. So far I haven't got enough time to include it into the algorithm to check the practicability of its use. Now I am interested most of all in the possibility of creating trading systems that will completely refuse from oscillatory indicators. In other words, is it possible to make market forecasts only by means of graphical drawings without using MACD, OsMA, Stochastic oscillators?
Concerning the criteria of breaking of a linear regression channel I also have so far purely visual observations that before a linear regression channel breaks a parabola top is formed. That is the trend reversal point can often (but not always) be approximated by a parabola having the peak in the trend reversal zone. So far I haven't got time to put it into the algorithm to check the expediency of application.
When my calculations showed the same residuals scatter for some areas for both LR and parabola - this was the main confirmation of the correctness of my algorithm for calculating these values.
It is not difficult to visually catch the top of the parabola when the LR channel is broken, but it is more difficult to teach the program. That is why the criterion I mentioned may be useful. The deflection between the centres of the LR and parabola, normalized to the scattering (as an option) can be involved. Not checked yet.
Strangely, I found the algorithm for teaching the program to identify the vertex to me immediately obvious. If A>0, then branches of parabola go up, then to determine that the top is already passed, you can use the condition Yparabola_current>Yparabola_previous. Conversely, if A<0, the branches are descending, and respectively the vertex is passed at parabola_current<Үparabola_previous. I look for a parabola satisfying the convergence conditions within 300 bars, for example.
I haven't tried to insert it into the algorithm yet, but the fact that this algorithm shows the passing of a top is visually obvious. I just have a linear channel(s) and parabola(s) displayed on the chart.