a trading strategy based on Elliott Wave Theory - page 44

New comment
 
Here's another surprise today. My algorithm (which I have not yet changed) calculates channels from 45 to 1000 bars. All channels that meet the CKO23>SCO criterion are counted. I assumed that these channels will be interspersed with channels that do not meet this criterion. I was supposed to find the best channels in each series and save them. When I change the sign of CCO23-SCO, the series counter is incremented and the series boundaries are memorized, then the best channel in the series is found inside the series boundaries. But this algorithm appears to have one important disadvantage - the series with a positive RMS23-SCO may not break for 1000 bars. Anything that cannot happen in the market is bound to happen. I tried to recalculate the morning channel on GBPJPY H1 and did not understand - the script does not work. It's a mystery! I have tried it several times before I suspected it. Checked it and it was confirmed.


 
Tried to recalculate the morning channel on GBPJPY H1 and didn't get it - the script doesn't work. Mystery! Ran it a few times before I got suspicious. Checked it and it was confirmed.

I also tried now to calculate GBPJPY H1 on 1000 bars. Got two linear regression channels on bars 63 and 164.
I calculate the sample by average bar prices (O+H+L+C)/4.
 
Yesterday I got some nonsense using my algorithm for pound pairs - I had to urgently introduce channel and zone building control - otherwise EAs just hung up with "unlisted error" and slowed the terminal. However, the Expert Advisor stopped responding to my surprise and, at least, did not freeze. That is, are considered equally bad (or good) and there is no possibility to choose - still perplexed ??? For the other pairs everything works. It's kind of weird. The algorithm is not related to the script given.Yeah, let's see.......

Good luck and hitchhiking trends.
 
Yesterday I got some nonsense using my algorithm for pound pairs - I had to urgently introduce channel and zone building control - otherwise EAs just hung up with "unlisted error" and slowed the terminal. However, the Expert Advisor stopped responding to my surprise and, at least, did not freeze. That is, are considered equally bad (or good) and there is no possibility to choose - still perplexed ??? For the other pairs everything works. It's kind of weird. The algorithm is not related to the above script.Yes, let's see....... <br / translate="no">


Still, I suspect a dumping into an infinite loop (something like this) with an unsuccessful channel search.
Also - this non-standard behaviour of the pound - perhaps this is some kind of signal when it becomes too good for this algorithm?
 
I don't think you're doing it right. <br / translate="no"> 1. The Hurst figure refers to a statistical series of numbers. It has nothing to do with LR or parabola.
2. Price is USD/EUR and Time is seconds, minutes, hours, etc. What you will add to what. The R/S ratio in the Hearst figure is dimensionless because R and S have the same dimensionality. And that is the only reason it makes sense.
3. In the formula H=Log(R/S)/Log(N/2), the value N is not time, as you might think. N is the number of elements in the sample. The fact that we do not take all events, but only a part of them (counting by Close, for example), dividing the process into equal time intervals, is our problem and it has nothing to do with Hirst.

Indeed you are right. Taking time into account in this parameter is probably nonsense!
So far I stopped at the initial calculation of the curve length as a simple price difference between neighboring points of the parabola. Subjectively I like such calculation better than the difference between High-Low samples so far. No final decision has been made yet. We are conducting experiments. I think that it will take some time of observations in order to understand to what extent this way of calculation R has the right to exist.
 
Another yesterday's channel that still holds up.

 
Still, I suspect it's going into an infinite loop (something like that) with an unsuccessful channel search. <br / translate="no"> And also - this non-standard behaviour of the pound - perhaps it is some signal when it becomes too good for this algorithm?


I think I've managed to replicate the problem on the history and fix it at the same time. The problem has disappeared as soon as I replaced Lowest(); and Highest(); with my own algorithm for searching. By the way, these functions are not suitable as it is not clear what exactly they output - the longest/lowest function value on a given interval (which is not appropriate) or an extremum (which is required), which is not the same thing - there are values at the ends of the interval and they can be more extreme than extrema but are not extrema - then the samples may be incorrect.

Good luck and good luck with trends.

SZZ I think that you will have to write all functions by yourself to be sure that the algorithms work and it will be easier to rewrite them on VCPP afterwards. :).
 
<br / translate="no"> I think I managed to repeat the history problem and also brushed it off. The problem is gone as soon as I replaced Lowest(); and Highest(); with my own search algorithm. By the way, these functions are not suitable as it is not clear what exactly they output - the longest/lowest function value on a given interval (which is not appropriate) or an extremum (which is required), which is not the same thing - there are values at the ends of the interval and they can be more extreme than extrema but are not extrema - then the samples may be incorrect.

Good luck and good luck with trends.

I think I will have to write all the functions by myself to be sure the algorithms work and it will be easier to rewrite them on VCPP. :).


"RE Slawa - answer to zigzag :)"

last post
 
It means that these functions give maximum/minimum values in the interval - it is not good in this case: the sample may include bars not related to the given trend which worsens (at least it does not improve) predictability. By the way, even Murray may sometimes draw levels incorrectly because of it - such an issue was....... Ok, I think everybody understood what we should fix.
To search for Hai, the condition should be at least

if((H[k-1]<H[k])&&(H[k]>H[k+1])&&(H[k]>curHi)) for minimum the same way.

Good luck and hit the trends.
 
if((H[k-1]<H[k])&&(H[k]>H[k+1])&&(H[k]>curHi)) for the minimum is similar.

All that remains is to justify the choice of a sufficient area around the High to take it as the High of the given sample. I take it that this High must be the maximal point within a radius of +-30% of the sample length? In case it is not so, the sample must be increased in order to determine two things together - the extremum and the sample length? What are your thoughts on this?

Vladislav, do you suppose you will correct the code of the Murray indicator in the light of the new information? We are waiting for the new version ;o)!
1...373839404142434445464748495051...309
New comment
Reason: