a trading strategy based on Elliott Wave Theory - page 269

New comment
 
Grasn, thank you very much! I would only discuss it after I've got into the subject, implemented it in the algorithm, tested it...

The topic will be relevant as long as there is a market.
(except that the fighters are all dispersed on conquered lands - not all reach the capital.))).
 
Didn't Vladislav mention this work somewhere in the bowels of the thread? The thing is, I read it a long time ago, got it from a spider, but the tip was from this particular thread I think. And thought everyone had read it :). My summary is this: The work is interesting and a few ideas based on it were. But in its immediate form such a model is too heavy for realtime (in the sense of consumption of machine resources). This means that it may be used only for large timeframes. And this means that it cannot be used to gather statistics on history, that is, in fact, it cannot be checked in advance.
 
to Candid

<br / translate="no"> Didn't Vladislav mention this work somewhere in the bowels of the branch? The thing is, I read it a long time ago, took it from a spider but the tip was exactly from this branch I think. And thought everyone had read it :). My summary is this: The work is interesting and a few ideas based on it were. But in its immediate form such a model is too heavy for realtime (in the sense of consumption of machine resources). This means that it may be used only for large timeframes. And this means that it cannot be used to gather statistics on history, that is, in fact, it cannot be checked in advance.


Quite possible. But I've never seen it (my absent-mindedness is probably to blame :o), but I've found it only recently and completely by accident. That is, I do not pretend to be the first treasure hunter. And I don't know why my colleagues bent rods and stretched strings... :o)))

Had to come up with my own PE, and by the way, works great. Indeed, the model is demanding on resources. Just don't understand why you can't get the statistics? I'm testing my simplified model now, for example, quite simply. I go in increments of +1 "into the future" and estimate the forecast. The history is quite enough. And I don't have any parameters either.

I don't really understand your logic. If it works, who prevents investing part of the deposit in a good server? Besides, you don't have to do everything on MT.

To Tovaroved


Grasn, thank you very much! I would discuss it only after I get into the subject, implement it in the algorithm, test it...


Absolutely no thanks to you, and not to me either. There's so much stuff on this internet ....:o)
 
I don't really understand your logic. If it works, who prevents you from investing part of your deposit in a good server?

As soon as it comes to a lot of money, logic is turned off and bare psychology remains.
This is a kind of fear of the unknown.

for example, for the last 8 months i have been writing a system (mts); objectively i have a week of work left, but it will take me much longer: from 2 weeks to 6 months.
the problem is that at the end of the work began to shut down the brain - I have to work on myself.

man is as inert as that material point.


Absolutely nothing to thank you for, and not me either. There are so many things on this internet ....:o)

for the search and publication.
the author does not read here - I thanked him mentally.


P.S. "thank you" is short for "thank you God!", so I don't recommend answering "for nothing". :)
 
I.e. I don't claim to be the first treasure hunter at all. And I do not understand, why colleagues bent rods and stretched strings... :o)))

Yes, I'm just protecting myself from reproaches such as "why didn't you say anything" :)
I just don't understand why you can't dial in the statistics? I'm testing my simplified model now, for example, quite simply. I go in increments of +1 "into the future" and estimate the forecast. The history is quite enough. And I don't have any parameters either.

I think it's a useful, but indirect method of checking. The direct method is history trading, and the number of deals must be at least 1000. However, maybe I'm being overly vague :)
I don't really understand your logic. If it works, who prevents you from investing part of your deposit in a good server? Besides, you don't have to do everything on MT.
Basically C seems to be 17 times faster than mq4, so there really are reserves :).
I must confess that my interest in "cheap" automatic trend building ( "Determination of trend start" ) was substantially caused by this very work :). However, this type of model still has a lot of unacceptably "heavy" moments for the frontal implementation. Well, I've already written that maybe I'll come back to the "potential" approach. But now another idea is in the foreground :)
 
to Tovaroved

<br / translate="no"> P.S. "thank you" is short for "thank you God!", so I don't recommend answering "you're welcome". :)


Thanks, will keep in mind :o)

to Candid


Basically C seems to be 17 times faster than mq4, so there really are reserves :).
I must confess that my interest in "cheap" automatic trend building ( "Determination of trend start" ) was to a great extent due to this very work :). However, this type of model still has a lot of unacceptably "heavy" moments for the frontal implementation. Well, I've already written that maybe I'll come back to the "potential" approach. But now another idea is in the foreground :)


Here I agree with Sergey (Neotron), it is almost impossible to determine the trend of such series statistically reliably, at least I don't know such technologies. In price series there are seldom areas where the trend is preserved by known methods and even if it happens it will most probably be by chance. It's pretty funny (from the height of the research I've done) to build a trend on almost (that "almost", changes a lot) Brownian motion (or noise, whatever). I may be wrong, though. I, for one, determine the strength of the relationship between the elements on the basis of correlation. That's where further calculations start. I've already bragged about my achievements. :о)
 
to Candid

About the trends. I decided to propose you a problem. Here is the "trend", eurusd, (H+L)/2. If this data is not enough, I can send you the full information or tell you from what period the data was taken. Yeah it really doesn't matter and it's not necessary :o)))).



I give you as a tool what is available and works perfectly - Hearst's indicator. You seem to know how to use Hearst. Let me just remind you how Hearst is calculated (instead of PE - Hearst). The value of the dead zone is 100. Hence, the Hurst value in 400 counts corresponds to the dead zone itself.



And here is the Hurst itself, already filtered:



can you predict the general price movement, will the trend continue (counts [0;500]) or will it reverse?

PS: If that's not enough, "system ripple" is attached. So far the working title is this, which is what I'm working on now. A couple of pulses are clearly visible... :o))
 
grasn, I don't think there's enough data to analyse. you need as much history as possible, then you can do some handheld tehanalysis. :)
 
2grasn
I guess I'll have to confess. I don't know how to use Hearst to predict prices :)
More seriously, I agree with Tovaroved, you need a longer backstory for predictions.
Also I think the correct way to set the problem is this: there are 1000 price chart segments, chosen according to some algorithm. We should correctly predict continuation of e.g. 80% of cases. The 80% figure is honestly taken from the ceiling, but it's clear that it should not be too close to 50% (spread and all that).
 
I agree, the task is not quite right, and there is not enough data. But it is quite sufficient for Hearst. And it is quite simple to use:

(0<H<0.5) High probability that the "movement" will change its direction
(0.5<H<1) High probability that the "movement" will keep its direction

We can see that channels up to 300 counts have values much lower than 0.5 and therefore are more likely to change their "movement". I just choose linear regression as an estimate of the movement. The ripple also says that the "live" spot on the charts is after 200 counts (approximately).

One can draw a simple conclusion that the "trend" [0;500] will change its direction, only the channels closest to the 500 count will remain for some time. This is how it happens if we observe the future.



And here is the very, specifically chosen place on the general chart:



This is what I am amusing myself with the whole history. Testing the components and the model :o)))

And I wrote all this just to attract your attention to the old Hearst when analyzing the trend. Very useful thing. The thing is I tried a lot of methods and found nothing worthwhile in any of them. Hearst gives very good results. Except, Hurst has to count right.

PS: I don't quite understand your problem statement. Why would I want to calculate the continuation of motion for 80% of the cases? And one percent of the cases is how many pieces of different motion? :о))))

500 count on the chart corresponds to 09.01.2004 (19:00), clockwise. The data is either alpari or metaquotes. Can't remember anymore.... Hurst is pretty consistent across different DCs.
1...262263264265266267268269270271272273274275276...309
New comment