
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
Using the indicator built in MT4 will automatically mean that you select the moving average as the forecast price. You can select something else. The algorithm for calculating the RMS itself is correct: the square root of the sum of squares divided by the number of degrees of freedom.
Good luck and good luck with the trends.
Vladislav, I would also like to clarify if I understand your recommendation correctly.
Let us give you Taylor formula:
Consider derivatives of parabola f(x)=Ax^2+B
f'(x)=2Ax,
f''(x)=2A,
f''(x)=0, all derivatives from third and above turn to 0.
Then according to Taylor formula we have series consisting only of first three terms. In this case the expansion of the function f(x)=Ax^2+B in the Taylor series will be exact (i.e. the last term for the expansion error turns to zero). Next, we need to evaluate the quality of approximation of the price series by the optimal parabola. That is, our main requirement is that the series of errors of approximation must be convergent (i.e. the sum of errors converges to a finite number). And we can determine this simply by comparing the calculated error of approximation with the third term of expansion. Am I right or not? So when choosing a parabola and the sample itself, we use the criterion that the RMS of approximation errors should not exceed the value of the third term of the series, for a sample of values lying in the interval from a to x? Do you follow the same principle in your strategy or not?
By the way, there is some inconsistency in this. We optimize the parabola using the property of price potentiality (through perpendicular to the parabola) and estimate approximation errors in the usual way.
What is wrong here? How can we reconcile finding the optimal parabola and estimating the approximation error?
Good luck and good luck with the trends.
A mozet sdelajem v all vmeste konstruktivnuju rabotu?
Say, napisat' sovmestno indikator, katoryj beget 4erez vs vs istoriju do teku4ej ceny i s4ityvajet Elliot waves :)))
Developery MT4 tol'ko pablogodorit za takoje.
Neskol'ko moix idej dlia na4ala:
1) At the very beginning istoriji opredelit' v kakuju toru cena ili FLAT
2) if FLAT, zdiom poka probivajutsia granitsia flata, tokda smotrim v kakuju storonu dvigajetsia cena, tak opredelajem na4alo ods4iota, s4itajem tol'ko 1-2-3 i A-B-C volny
3) is4em tol'ko "basic" Elliot Wave patterns 1-2-3 i 1-2-3-4-5 + A-B-C volny after okon4anija dvizenija ceny(trend)
4) Jesli imejem "failed Elliot Wave", zna4it ploxoj ods4iot i tot kusok istroriji nada jes4io raz peresmatret' nas4iot v kakuju storonu dvigajetsia cena intervale pabolshe teku4evo.
5) K etim grafikam xorosho godosho cifra Fibonacci, sami lookotrite s indikator MT4 in istoriji from Elliot Wave 1 na4ala to na4ala Elliot Wave 4 - http://www.market-harmonics.com/elliott_wave2.htm
Dopolnitel'no doli poniatija o 4iom re4' pro4itaite http://www.elliottician.com/showpage.asp?p=47 i postaraites' ponat' kak kotritsia "bassic Elliot Wave pattern". Polnoje opisanije na ruskom ses' : http://www.alpari-idc.ru/ru/textbook/tech_an/ew/
V rezultat kod indikatora patom mozno podkrutit' k novojiji versi MT4 kak standartnyj indikator :)
So you probably do the following.
Step 1. Take a sample
Step 2. Approximate it with a linear regression channel
Step 3. Find approximation errors.
Step 4. Analyze the graph of errors. We get the supposition that the order of approximating function should be higher or the given sample cannot be approximated by any continuous function at all if the error series diverges or has some strong deviations visible to the eye which fall out of the acceptable confidence interval(the calculation automation algorithm is not completely clear yet).
Step 5. Repeat steps 1-4 for approximation by a parabola (or something else)
Step 6. Evaluate the errors; if the errors exceed a reasonable limit, you simply discard this sample. If the error graph has some reasonable structure, then we store the information about sampling, approximation method and additional information about approximating functions in some array.
Step 7. Then, having repeatedly tried all possible samples and having searched for the optimal variants of approximating functions for each sample, we stop at those samples which satisfy our requirements in the extreme way. It is also naturally desirable to use your recommended method of approximating functions not for the entire sample, but only for 2/3, leaving the last third to test the approximation results (this is a very valuable suggestion!).
Step 8. Draw extremum approximations on the price chart with continuation into the future. It is natural that a confidence interval is plotted for each approximation.
Step 9. Thus we see where the limits of intervals intersect. Then we define the approximate dates.
Step 10. During the price approach to the turning points, we calculate the probability of trend reversal using the integral error estimation method. It will probably be necessary to average the pivot estimates for all approximation channels. For the linear regression channel it will also be necessary to calculate the Hearst coefficient to have it as an additional parameter. It is also good to look at the Murray levels. Thus, we have a high probability of making a decision about placing pending orders and determination of stops with minimal risk.
Of course, the Expert Advisor that will calculate all that will be very extensive (you said it contains 6000 lines)! And so far not everything is clear in terms of automatic decision making for each of the samples. Well, I think you just need to start trying to program this algorithm, and then as you experiment, you can figure out something that is difficult to understand even on a theoretical level, but that will become clear by itself during the experiment. And indeed the calculation time will be quite significant. You said that the first variants worked for 30-40 minutes on a weak machine. Well then on a P4 2.4 GHz you should expect about 10 minutes computation time.
On the subject of approximation methods I found the following interesting tutorial.
Since you don't need the parabola itself, you can approximate the derivatives right away. The regression coefficient is what you need (hence the Taylor series ;) ). Then you won't care what the shape of the trajectory is - the main thing is to correctly estimate the confidence interval. Please read carefully the recommended literature, it contains enough information.
Good luck and good trends.
Got it... :)
Vot odin iz moix staryx mql3: