a trading strategy based on Elliott Wave Theory - page 9

New comment
 
<br/ translate="no"> Generally speaking, it is not a secret - it depends on the level of risk that the trader is ready to accept. For example, from January till the end of February I have almost doubled my deposit on real account (more exactly 93%), trading with minimal risks. And for monitoring I have posted a demo with maximal risks - there I traded almost 450%, but I would not risk so much on real account :) - These were the best figures, while the average was about 40%.

In my strategy to get such a value as 93% profit over 2 months one should set maximal drawdown at 50-60% in money management. Of course I can not allow that on the real market.
Your strategy probably deserves attention with such parameters. But the information presented is not enough for me to try to develop an Expert Advisor using your strategy. Basically, if you believe that the strategy is not subject to publicity, I can understand it. But if you still want to present it in a practical application, maybe someone will want to automate it (maybe even I can take something useful for myself from it).
 
Hi Vladislav!
Thanks for your openness and interesting posts ! I like the logic and basis of your approach.
Could you please give some link to Hirst's criterion?

You touched on it very briefly in your posts and I, alas, know nothing about it. However, the topic - estimating
It is interesting for many people, including those on this forum.
If possible, not abstract mathematics, but something more or less practical,
from which one could understand both the idea and how it counts on counting sets.

And another question.
The result is as follows - we get not only the Murray levels from which the pivot points are taken, but also their statistical significance at the given moment in time.

Is this your methodology or something open source?
Maybe you could say a little more about it ?
 
You can find everything here if you want to :) :
http://forex.kbpauk.ru
 
You can find everything there if you want to :) :<br / translate="no"> http://forex.kbpauk.ru

Yes, maybe, but I'd like a recommendation from someone who understands what he recommends.
Why would I, new to the subject, want to dig through a mountain of rubbish where the vast majority are
are newbies like me? Waste of time.
 
Ok - no problem. For the Hurst criterion, you can start with the spider - there's a starting point there. You can also find it on the search engine. The criterion itself gives an empirical estimate of how far we are from a distribution converging to normal (that's just to put it simply). That is, an estimate of convergence with increasing degrees of freedom in the sample. It is also associated with the estimation of the fractality of the market (in the Mandelbrot sense - not to be confused with D'Billiams !!!! ). There is also the central limit theorem which says that any convergent distribution with increasing degrees of freedom converges to normal distribution (therefore it does not really matter what it is inside, as long as it converges ;) ), and this means the applicability of the mat statistics tool on samples longer than 30 degrees of freedom (bars in this case, but not on any sample !!!!!!). - Therefore the algorithm turns out to be iterative) - errors there will tend to zero - hence analysis of small periods by such methods is doomed - I think so. When I calculate daily levels on intraday charts the error is insignificant - the sample length is enough. So this characteristic when approaching 0.5 - means that white noise prevails, and when approaching 1 - indicates the presence of stable structures, and 0 - unstable. Further interpretation in applicability (mine, though probably obvious): stable - trend, unstable - counter-trend. A fractal in this case (i.e. self-similar structure) is a regression channel, of course with target estimation - otherwise it is simply infinite and there are problems in use ;). Generally the method (also non-trivial task) reduces to search for channels - there can be more than one at a given point and most importantly - to choice of the most suitable or a superposition of suitable if there is more than one. Their boundaries outline the turning zones in terms of price and time. In fact, at each bar the optimization problem is solved. The calculation method itself takes almost 0.5M in codes - compare with the size of your indicators.)
Yes, once again I remind you - all this makes sense within the problem I have formulated and the result is interpreted within the same problem. Although the methods are common. As for the statistical significance of levels - it will become clear to you once you build confidence intervals - for example, the more oversold, the more likely it is to return to the equilibrium point and possibly even to the opposite boundary (this becomes clear when you approach the equilibrium point) - so the confidence intervals cut off the probability levels. Imagine that you numerically interpret the oversold area in units of probability returning to the equilibrium point (in percents - hence the probability of returning 60\40, hence 80\20, etc... ;) ) and the reversal levels by Murray at a given time fall into, for example, 90\10 - it would be easy to trade? And there is less ambiguity, isn't it? So that's the solution to this problem that gives such an estimate.
At reversals all these structures line up, well just a peek - then the probability is maximal. I traded in demo mode without any stops.) I'm not risking it on the real account, although I'm tempted to open an account for some small price and try it :).

If you'll excuse me, I won't show you a ready-made solution, although I've saved you a lot of time.

Good luck and good trends.
 
Thank you Vladislav !
What is R/S ?
 
Thank you Vladislav ! <br/ translate="no"> What is R/S ?


Hello again ! R / S - statistics is a Hurst criterion (statistics) - his formula, if you can not find it on the web - write to us, I'll send it - R - the superposition of disparities, S - RMS (standard deviation). In the formula there is a logarithm of this ratio - that's why they call it this way and that way.
What else I forgot to pay attention to yesterday - and it's essential - two things:
1. Emergence of an optimization problem. Alas, without it I failed to achieve unambiguity - this problem arises from a hypothesis (the hypothesis must be taken into account when setting the problem) that the price follows the only possible path, along a trajectory which we do not know with 100% probability. Since the price field is potentially swap-precise (here the term "field" is used in its mathematical sense - that is, the function together with its area of definition, and the function is the trajectory sought ;) ) - Strictly proving it is not difficult : a potential field is that field whose work of forces on a closed loop (integral on a closed loop ;) ) equals zero - so "on the fingers" it looks like this - no matter where the trajectory goes up/down, but if you return to the starting point, then your earnings are zero. From this you can make an assumption that the trajectory function can be adequately represented by a certain quadratic form - it's almost simple: searching for extremums of quality criterion functionals for such forms is a highly researched area. That is, we need to select samples that extremally satisfy quality criteria.
2. If a methodology allows to "draw" some objective regularities, then the result should not be sensitive to "noise" - it is logically understandable. So, since January 2006 I have managed to obtain solution methods, which provide identical levels and reversal zone borders in any datafeed (available to me), i.e. in any brokerage company, despite some differences in quotes and I don't use smoothing algorithms - they all lag.
I really want to believe that the increased effectiveness of forecasting has something to do with it (it seems to be justified logically).
Although it may still turn out not to be so :) - anything is possible.
I will use the trading simulator from Scientific Consultant Services, Inc. (scicon) - hopefully it will help me to clarify the situation.

Now, it seems, that's all. Because we have rubbished another thread with unnecessary information :). Good luck.

2 Begun - if all this has long been known (I'm on a spider at the time I found only the starting points) - would you share information, at least in methodological terms, otherwise I may be wasting my time on constructing a bicycle ?

Good luck and good luck with the trends.
 
<br / translate="no"> R/S - statistics is the Hurst criterion - its formula, if you can't find it on the web - send it to me - R - superposition of deviations, S - RMS (standard deviation). The formula contains the logarithm of this ratio, which is why it is called this way and that way.

I searched the Internet. Could only find some software for calculating in Excel http://megafx.fromru.com/FRAGKTVBA.xls.
But it's not easy to understand, because I've never coded in Excel.
Please write the formula here. It will be interesting to many people. And also preferably a piece of code which you use to calculate this value.
 
<br / translate="no"> Please write the formula here. It will be of interest to many people. And also preferably a piece of code which you use to calculate this value.


It is strange, I have not found it in my archives - my CD has gone kaput :(. But I found it on the web - in general it is even more interesting than I initially thought:


1.1 Evaluation of the Hurst Index

Let us first outline the theoretical background to the quantitative mathematical analysis of exchange rates series below.

The Hurst H or, as they say, the Hurst statistic R/S, indicates the presence or absence of a bias in the series in question. In the RC such a bias is generated by market participants reacting with a bias to the current economic environment. This bias continues until new random information emerges and changes this bias in magnitude, direction or both.

Hurst's R/S analysis gives us two important characteristics of a time series. Firstly, the average cycle length required to estimate the inertia of the motion. The average cycle length of a system refers to the length of time after which the memory of the initial conditions is lost.

Secondly, the Hurst exponent is stable, contains minimum assumptions about the system under study and can classify the time series differentiating a random series from a non-random one, even if this random series is not Gaussian. For example, if the Hurst index differs from 0.5, it means that the probability distribution of the studied time series is not Gaussian. If 0 < H < = 1, but H is not equal to 0.5, then the series is a fractal whose behaviour is significantly different from random walks when H = 0.5.

Thus, if H = 0.5, the time series under study is Brownian motion, the observations are independent and have Gaussian distribution. But if H > 0.5, it means that the observations are not independent. Each observation carries a memory of all preceding events, and this is not a short-term memory called a "Markovian" memory. It is a different long-term memory, and in theory it is always retained. Recent events have a stronger impact than previous events. On a long-term scale, the system which gives Hearst's statistics is the result of the interaction of a long stream of interrelated events. What happens today affects the future. Where we are now is determined by where we were in the past. Time turns out to be a very important factor here.

The most important applications of the Hurst index H are the following.

If H = 0.5, then the Efficient Market Hypothesis (EMH) is confirmed, i.e. yesterday's events have no effect today, and today's events have no effect on the future. Events are uncorrelated and have already been used and depreciated by the market.

In contrast, with H > 0.5 today's events will matter tomorrow, i.e. the information received continues to be taken into account by the market some time later. This is not simply autocorrelation, where the influence of the information drops off quickly, but it is a long-term memory, which conditions the information influence over long periods of time. Of course such influence does diminish over time, but it is still slower than short-term correlations. This influence is characterised by the length of the cycle, when it drops to an indistinguishable value. In statistics it is called the series decorrelation time.

Thus, if the fractal nature of the time series is proved, it means that the Fractal Market Hypothesis (FMH) is proved, which in its turn contradicts the GER and all quantitative models which are derived from this hypothesis.

To quantify H Hurst derived an empirical law in the form:


H = Lg(R/S)/Lg(n/2)
R - maximum range of the studied series
S - RMS (standard deviation)
n - number of observations (sample size)



The codes will hardly tell you anything: there too many connections, as all arrays are filled in corresponding places - I've already written the procedure is iterative, but nevertheless:

 
  //--------------- Hurst coefficient double R = 0.0, pMax = 0.0, pMin = 0.0, S = 0.0, nHrst = N_BG[i_StdChnl][1]-N_ND[i_StdChnl][1]; if(nHrst>minChnlBars){ S = std_div[i_StdChnl][1]; pMin = Low[Lowest(NULL,0,MODE_LOW, N_BG[i_StdChnl][1] ,N_BG[i_StdChnl][1]+StepBack)]
       pMax = High[Highest(NULL,0,MODE_HIGH,N_BG[i_StdChnl][1], N_BG[i_StdChnl][1]+StepBack)]; R = MathAbs(pMax-pMin); if( (R>0)&&(S>0)) Chnl_Hrst[i_StdChnl][1] = MathLog(R/S)/MathLog(nHrst*0.5); }



Good luck and good luck with the trends.

 
To summarise your strategy for myself (translating it from mathematical language to engineering/algorithmic language that I understand), I would like you to confirm whether I understand it correctly or not.
You have the following calculation modules (or parts of the strategy) in your strategy:
1. Calculation of Murray levels (this is in principle clear enough in terms of implementation, the more so, as the indicator itself is given in this thread).
2. Calculation of a regression channel that includes know-how (the criterion for selecting a correct channel of the set of possible ones), which you are not going to share.
3. Calculation of the Hurst index for the sample, which is determined, again, by some criterion that you are not going to share with the public either. Or may be I am mistaken and you just count straightforwardly for example by the last few bars? Then once again, specify the number of bars and time frame. The number sounded like 30, but maybe you are using other values.

And then you get the calculated data from the above modules and make the following conclusions about the market. I have shown all the variants figuratively:
1. The market is increasing. It is now near the Murray line, which implies a stop and reversal. The price is in the upper part of the regression channel, the Hearst indicator is moving towards zero. Conclusion: It is OK to enter short. A stop is placed behind the next Murray line, which is strong resistance. The initial target is the nearest strong support line. Then based on the readings obtained from the calculated modules when approaching the target, we decide to keep the position or close it, if the indicators recommend to do so.
2. The market is falling. It is now near the Murray line, suggesting a stop and reversal. The price is in the lower part of the regression channel, the Hearst indicator is moving towards zero. Conclusion: It is possible to enter into a long position. A stop is placed behind the next Murray line, which is strong support. The initial target is the nearest strong resistance line. Then based on the readings obtained from the calculated modules when approaching the target, we decide to keep the position or close it, if the indicators recommend to do so.
3. The market is in flat. We make assumptions about further continuation of the movement based on the readings of the Murray line. If we have an open position and its direction coincides with the line readings and with the Hurst Index (for example, when the predicted movement coincides with the position, the indicator is close to 1 or 0), we do not take any actions with the position, and wait for the objectives to be achieved.
4. The market is in flat, the Hurst reading is close to 0.5. We do not enter the market; all the orders are removed. If we want, we may use pipsing.)

Am I right in my reasoning about your strategy or not?
12345678910111213141516...309
New comment
Reason: