a trading strategy based on Elliott Wave Theory - page 33
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
Just a general question.
Yuri, as for the practical implementation, or rather the methods underlying it, everything is quite simple: the quadratic function has coefficients which you need to select in an optimal way - regression gives a linear, or rather an estimate for its construction. And, accordingly, you will be able to estimate up to what limits (amplitude spreads) in the Taylor expansion (construction of the quadratic form) this coefficient can be used. Further, as for other coefficients, think for yourself. And to find minimum of potential energy you don't need to know price trajectory, but what is more important to know - potential gradient ;). That is, the dynamic state of its zero-potential - you have to count something for the zero-potential. And all this is sufficient to estimate - direct differentiation is not necessary.
If figuratively, "on fingers", applying geometric images:
just imagine that on the surface (analog of some rugged terrain) a ball rolls (this is the price). It is not necessary to know the intricacies of the ball's workmanship in order to determine the areas of attraction of the ball's trajectory. It is much more useful to know the properties of this 'rugged terrain'.
I think that's really the end of it - the comments are over.
The above is sufficient to build, if not exactly repeat, a similar strategy, and comments solandr quite confirm it :).
Good luck and good trends.
I read your last post. I'll say it again - just a piece of advice - don't try to look for/identify the spreads for the price trajectory. Approximation errors when there are many factors influencing the trajectory and if the influence of these factors is about the same, if with increasing degrees of freedom (order of approximation, dimension of samples) converge, they converge to a normal distribution and this is a proven fact, use it ;). And this means that confidence intervals for them (errors) can be estimated more reasonably than for the trajectory itself. (Much if in one sentence, but I can't formulate it any simpler...) Otherwise you risk not seeing the forest for the trees.
Did you mean to say a sequence of one-bar "tails" from approximations of sub-samples by a function of the selected form? Still, it's a slightly different approximation. ;-) And we'll get an answer to the question of how much this sequence of "tails" resembles the driving force, rather than the original function.
Have I got it right?
Thank you for your recommendations and advice.
Unfortunately we, for some unknown reason, cannot understand each other.
Everything you wrote in this post is indeed a repetition of what you have said more than once. However, I didn't need to repeat it. I appreciated the strength and coherence of your approach from the very beginning (as soon as I realized :-) and expressed my admiration for you back on page 5 of this thread.
But I am not trying to repeat your approach, I am not looking for distributions and I am not differentiating anything.
I am not able to do anything at all without understanding what I am doing. The discussion in this thread has drawn my attention to some points (e.g. - probability estimation) that I hadn't even thought of before. And this is quite useful for my own system. That's why I'm trying to understand some technical details and participate in the discussion.
You do build confidence intervals using cramps, don't you? And their width in sco units is unambiguously determined by the distribution. So, I asked how you do it precisely because I'm not going to look for it and I'm satisfied here (for want of mine) with someone else's experience.
As for calculating the Hurst Index, I do not understand your unwillingness to comment on it. A simple technical point. Nothing to do with the subtleties of strategy or secrets of your methods. In order to get a meaningful result it is necessary to have a meaningful calculation algorithm. I was taught to fold apples and pears at school :-)
What you've written about approximation errors, I understand it and use it. Thank you.
In principle correct, but only when constructing the confidence bounds of the channel the action, or direction of that very driving approximating function will obviously extend over 1 bar if it is indeed the function that is driving the price right now and not a function that is picked up at random. It is quite understandable that in calculating the channel, the channels themselves will "stray" within some limits. Well we are trying to understand the interdependence of each successive transaction on the previous one, aren't we! (See the same book.) And to understand how deals are interdependent we can take into account those factors which influenced exactly at that moment in time (which channel or pre-channel of the main sample existed at that moment in time), when the previous data were known and based on which the crowd was making an average decision for the next bar, so to speak.
If you're worried about the fact, that the next bar can break the channel, in fact, if there was not some extreme event, but some standard news, which the crowd was waiting for to go where the price should go, then the next few bars will remain in the same channel approximated by this function. If the news turns out to be the wrong news, the market reaction is usually quite sluggish. In such cases the news reports are of the type "The market has reacted weakly to interest rate hikes". In cases when the news happens that nobody expected, the crowd may rush to the wrong place (not in the right channel), but after a short time, the crowd just does not know what to do next. And only after some time, when new channels start to form, the crowd starts to act more consciously. And very often, the price tries to return to the channel, from which it accidentally fell out due to unexpected news.
So, I kind of wrote how I do it - by the confidence interval. Let me explain in more detail: if the regression channel describes the movement correctly, then ideally all prices should lie on the regression line. What does the width of the confidence interval mean? That with a given probability the trajectory will lie inside. If the channel is correct and we have nailed one of the bounds, it's easy to recalculate the probability that we will go back.
You do build confidence intervals using cramps, don't you ? And their width in sko units is unambiguously determined by the distribution. So, I asked how you do it precisely because I'm not going to look for it and I'm satisfied here (for lack of mine) with someone else's experience.
Yes. I do it simply: I compare it to the "worst" distribution that still converges.
As for calculating the Hurst index, I do not understand your unwillingness to speak on the subject.
As I've already said many times before, after identifying the sample I calculate the Hurst index for the selected channel to make a conclusion about the way it (the channel) participates in the forecast.
Good luck and happy trends.
Po4itav o 4iom vy sdies' govorite s kooficientom Xersta ja pdumal 4to mozet vam prigoditsia i moja narabotka kokda tryu opuoznat' ods4iot voln na UP or DOWN or flat. Skinu fragment from svojevo indikatora (u menia priviazka s Fibonacci golden ratio):
extern int HighLowPeriod=350; extern int PriceShift=0; extern double TrendFactor=0.38197; int start() { int MaxPriceBar=0; int MinPriceBar=0; double MaxPrice=0; double MinPrice=0; int WaveAngle=0; MaxPriceBar = Highest (NULL,0,MODE_HIGH,HighLowPeriod,1); MinPriceBar = Lowest (NULL,0,MODE_LOW,HighLowPeriod,1); MaxPrice = High[MaxPriceBar]; MinPrice = Low[MinPriceBar]; //additional conditions for WaveAngle goes here if (MinPriceBar < MaxPriceBar) // Counting UP { if (WaveAngle == 0) { if ((Time[PriceShift] - Time[MinPriceBar]) < (Time[PriceShift] - Time[MaxPriceBar]) * TrendFactor) WaveAngle = 2; // Possible DOWN trend if ((Time[PriceShift] - Time[MinPriceBar]) >= (Time[PriceShift] - Time[MaxPriceBar]) * TrendFactor) WaveAngle = 3; // Possible trend end } } else { if (WaveAngle == 0) { if ((Time[PriceShift] - Time[MaxPriceBar]) < (Time[PriceShift] - Time[MinPriceBar]) * TrendFactor) WaveAngle = 1; // Possible UP trend if ((Time[PriceShift] - Time[MaxPriceBar]) >= (Time[PriceShift] - Time[MinPriceBar]) * TrendFactor) WaveAngle = 4; // Possible trend end //if (Symbol() == "EURUSD") Print("WaveAngle:",WaveAngle, " MaxPriceBar:",MaxPriceBar," MinPriceBar:",MinPriceBar," (MinPriceBar - PriceShift) * TrendFactor:", MathRound((MinPriceBar - PriceShift) * TrendFactor)); } } Comment("WaveAngle:",WaveAngle); }Eto davolno prastoj metod, no. o4en efektivnyj, kokda vash kooficient Xersta byvajet 0.5+, v etom kode dumaju WaveAngle budet imet zna4enije 1 or 2 :) Eto dajot verojatnost' 85%+ 4to ods4iot na4ala (UP/DOWN) voln Elliota - pravil'nyj.
Что же касается расчета показателя Херста, то мне совершенно непонятно Ваше нежелание высказаться по этому вопросу.
I have already said more than once that after identifying the sample, I calculate the Hearst index for the selected channel in order to make a conclusion about the way it (the channel) participates in the forecast.
Vladislav,
You must be joking, repeating what you already said more than once?
But if you're not and I was wrong, then I'll repeat for the third time the issue around which the discussion is actually taking place.
solandr, when calculating the Hurst index, uses the scales of approximation errors and counts the spread as High - Low prices (and not errors). Is it correct?
Saving your time. Just yes or no ?
Good luck.