a trading strategy based on Elliott Wave Theory - page 37
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
The only problem is the impossibility to get a value less than 0.5 on large samples.
I read a simple and clear piece today, about the Hearst index and its calculation, from which several things follow. First of all, you can't just count pX as a fraction of logarithms for the whole dataset at once. The denominator should be Lg(aN), where a is an unknown constant. Assuming a=0.5 is arbitrary. For the normal distribution a=pi/2. This is why we need to read out Lg(R/S) as a function of Lg(N) and then approximate this dependence by linear regression. Then H is the slope angle and the coefficient a is the free term of the regression. Even if a=0.5, this algorithm should yield different results.
Secondly, the whole theory is only applicable to a series of basic data, i.e. a series of prices, for example. It is incorrect to apply it to the series of linear regression errors (i.e. to the series from which the trend component has been removed). For such a series, neither the spread nor the slope (especially on a finite interval) is time dependent.
Regards - Alexander.
In 2 words "on bricks" to explain everything written in the books is probably impossible. I can only try to answer your specific questions.
By the word INDIRECTION we understand the "Soviet" concept of DIRECTION of the function, by which you approximate a price series to the series itself. That is, if the price moves around a line that resembles a parabola for example, then a parabola is the best CLOSE. If you like, you can think of the main trend as following a parabola. So in this case it is obvious that if the approximating function is a straight line, while the trend is clearly following a parabola, we can say that the straight line is a abnormal CLIMIT.
About 30 bars. I mean the minimum number of bars of the sample itself at which the calculation data of e.g. parameters of regression equations etc. that will be obtained on their basis will be worth something in terms of statistics (reliability of calculations). That is, if you take a smaller number of sample bars, then those parameters that you will calculate can be called parameters that are obtained by chance and cannot be trusted. By increasing the number of sample bars, the credibility of the parameters to be calculated increases. It should also be noted that for ANY number of sample bars, ALL parameters that will be obtained in the calculations will in turn have some variation in terms of credibility. That is you can use formulas from the book to calculate, for example, that coefficient a of the linear regression equation equals 5 plus/minus 1 with 99% probability. That is, it tells you that the parameter that you have calculated is not really equal to 5, but is equal in 99% of cases to the value falling within the range of 4...6 and only in 1% of cases a may take values beyond this range 4...6. And this range narrows all the time as the number of bars in the sample increases. There are formulas in the book that can be used to calculate this range of values, called the confidence interval.
I understand: if I conditionally divide the LR channel which is closest to something (from my right or wrong understanding of approximation), then the Confidence Interval is the point where the current price is found in relation to the channel width in % ratio or in other words "for example if we take the bottom of LR channel as 0 and the top as 1, the price is somewhere md 0.01< price<1"
Please explain, if something is wrong.
If you take the linear regression equation y=ax+b, then as I said above, each parameter in the equation has its own spread, which is calculated using the formulas in the book. As well as the parameter a, the parameter b has its own scatter (the interval in which it actually lies). That is for example b=10 plus/minus 3. It lies in the interval 7...13.
If you except the linear regression equation y=ax+10, plot two more equations y=ax+7 and y=ax+13, the area between the upper and the lower line will be called the confidence interval. The confidence interval (spread of parameter) for intervals with different confidence probability will be DIFFERENT! That is to say, without thinking about specific coefficients, I can give the following example on my fingers. Let's take the same parameter b=10. Then, for example, the probability that this calculated parameter really does lie in the range 9...11 is 60%, in the range 8...12 is 80%, in the range 7...13 90%, etc. Actually the numbers are taken from the ceiling - the correct values have to be calculated by formulas. So the point is, the more certain we want to know the parameter, the wider we should take the confidence interval. Accordingly, with a low probability we have a narrow range of values with a high probability - a wide range of values.
That is, the channel is plotted from the central regression line in both directions. And the probability is applied exactly to this symmetrical area regarding the estimated regression equation.
In general Vladislav actually meant exactly the same thing that is written in Peterson's book "Chaos and Order in Capital Markets". On a nutshell, the gist is as follows. The EU has after a certain moratorium started to raise the Euro interest rates. You can watch and see that from that very moment the Euro started to rise against the dollar. Although, the rate has not risen much, but overall traders had the opinion that "EUR should start to grow now", which has been circulating in their heads for 4 months. So, whether the traders want to realize it or not, they have been pushing the Euro for a long time, willy-nilly. That is, every subsequent trade increases the rate of the euro, although I think that 99% of traders would say that they have forgotten about what it was 4 months ago at all! Nevertheless, it WORKS! Well, just in recent weeks, the growth of the Euro has slowed down, as traders now really start to forget why they were pushing the Euro up for so long. And now the market is waiting for something new, which will give it direction. And it is better that it will be the news, which most traders agree with. Since the price series with different number of bars can be approximated by different functions, in addition to the global event that formed the long-term direction of the euro trend, there is a lot of local events due to which the price jumps around the main trend. Therefore, weak events have less influence, bigger events have a greater influence. Thus, the price movement will be caused by the sum of these influences.
By error we mean the error between the approximating function and the real price series. Of course, if you ran a linear regression on a sample that also contains a parabola, the error graph will show the same parabola that you did not take into account. And you will need to subtract the parabola from the resulting errors to estimate statistical parameters such as RMS, which is one of the coefficients when calculating the confidence interval.
For you it is better to write the equation in the following form Price=a*Time^2+b*Time+s
I can't explain it in more detail.
it's not clear what is the prefix: 1. just the price relative to the current bar to the right; 2. the upper/lower BAR OF the channel LI at the current bar or relative to the current bar to the right
The prediction limit is the area where the boundaries of confidence intervals of different channels intersect. That's where they intersect, that's where the price should turn.
The projection is a continuation to the right of the linear regression line and channel borders (upper and lower straight lines or curves parallel to the central line (see explanations above)).
Just consider the part about 2/3 as an axiom (as truth) and leave the rest out - it doesn't matter here.
You have confidence intervals constructed. At ANY point you can use the formulas in the book to figure out on the boundary of which confidence interval that point lies. For example, you point your finger to a point without looking. Suppose you missed the channel for which you want to estimate probability. And the channel was built for a level with 99.9% probability. So, relative to the point you missed and did not enter the channel at all, we can say that the probability of this point being in this channel is no higher than 0.1%. This means that if the point you have poked at had a real price, the probability of this case would not exceed 0.1%. Now, what should the price do next, when it reaches this point? Probably, it is not difficult to guess that in the shortest time it should have moved back to the channel. Then it is a matter of technique - look at the channel and your finger and place orders. And then everything happens by the same algorithm for the case when you hit the channel itself. Through the point in the channel of the linear regression you have clicked, you can draw a line that is the boundary of a certain confidence interval. And then you just have to use the formulas in the book to figure out which channel? What was the confidence interval. Suppose you calculated it and understood that the probability was 75%, then you can deduce that the probability of finding the price outside the channel is 25%, and in the channel - 75%. And you may draw conclusions about where the price may go next.
Actually, in terms of quadratic forms I proposed a similarity to the one used by Vladislav. He uses a quadratic form of the form F(x,t)=Ax^2+Bt^2+C and also uses a field gradient. He somehow either finds centres of these quadratic forms in the plane and the coefficients of the equation themselves, which allows him to easily determine fluctuations of the potential field gradient, from which he draws conclusions about the field potential (or rather its fluctuations). And it allows one not to pick up the parabola itself. I have seen in books about it, but how to apply it in our case yet I do not understand :o(. That is, the point is as follows. Imagine a terrain on which stand such elliptical cone-hills. These hills combine with each other in different ways. So the trend is moving in those places where one cone intersects another. How to calculate this I do not know yet. So far I have suggested a more comprehensible parabola equation for myself. And he does it in some other way.
Frankly speaking, I can't think of anything to add to what I've already said. The more so that there is a reasonable discussion going on in this forum that this indicator has no relation to the forecast. Well as they say everyone is entitled to their own opinion.
https://c.mql5.com/mql4/forum/2006/06/ang_error.zip
Unfortunately, mql4 forum is unable to directly attach gif files for some reason. I can not even imagine for what reason. I can only upload zip files to the forum perfectly. I used to paste gif pictures in the same thread without any problems, but since they moved to a new engine it's like the gif has been disabled, at least from my computer. Insert file C:\temp\ang_error.gif but the message is inserted without the file. Well, well, let's work as it turns out.
Well, EURUSD H4 shows that the coefficient in term X^2 seems to be taken with the opposite sign. How does it happen? Maybe, the algorithm makes such errors? Can we simply add an additional check of the indicator by MOC at two variants of the coefficient and display the one that simply shows a smaller error value on the chart?
The fact is that the trend along a parabola is very approximate. Rather it is a combination of a set of parabolas and rectilinear channels. If you try to extrapolate a parabola it may go sharply into "space" or corkscrew into "ground". Actually, in my opinion, a better, albeit more difficult, approximation based on wave methods, in particular based on decomposition of the trend into harmonic Fourier series. I understand VG. It's much easier to calculate and make line-based automata, but that doesn't mean it's better. Although, if done competently, the results are likely to be much better than if you try to use traditional methods, which are mostly used in technical analysis nowadays. Most of the methods of technical analysis were developed years ago in the slow markets, mainly for the days. What you can do now on the computer - I think Uncle Gunn would weep with happiness. So breathe deep and smile a lot.
I wish you successful trading and a quick realization of the uneasy business.
Regards - Alexander.
2006.06.05 12:07:54 ang_script EURUSDm,M30: invalid time value for ObjectMove function
Maybe something else should be done besides running it on a chart?
Try downloading the MT4 version from this server, and open a demo account.
Or even better bild pre194. I just checked, everything is working fine.
Rosh, in principle the derivation of the equations themselves is obvious. Everything is clear with it. But I take it that you are using averages for x and y. That is, you simply solve one equation by linear algebra methods. But what I don't understand is the following. Is it really possible to simply substitute sample averages into these formulas and get exactly what we need? Could you give a proof of this?
Does the ANG3110 indicator work according to this principle?
I think it would be more logical to solve N such systems for N bars and from the sample of obtained arrays a,b,c determine the expectation of each parameter and use it as a parameter for approximating parabola. Or am I mistaken?
Now everything is clear in terms of the solution!