a trading strategy based on Elliott Wave Theory - page 134

New comment
 
That's very plausible. If only we could find out exactly how the actual turnovers of this pair were changing. At least the assumption that EURUSD is the most statistically secured pair looks uncontroversial.
 
The other day I found an archive which is almost two years old. It contains code for MT3 and comments in txt-file.
I must be Dart.
/*[[
Name :=
Author := 
Link := 
Separate Window := Yes
First Color := Red
First Draw Type := Line
First Symbol := 217
Use Second Data := Yes
Second Color := Blue
Second Draw Type := Line
Second Symbol := 218

]]*/

Variables : shift(0),AccountedBars(0),FirstTime(True),i(0),swap(0),MinL(0),MaxH(0);
Inputs: PeriodHerst(24),PeriodA(100),CountBars(900);
Array: ArrayOne[250](0);
Variables : Average(0),Deviation(0);
SetLoopCount(0);
If FirstTime Then Begin
	AccountedBars=CountBars;
	FirstTime=False;
End;
For  shift = AccountedBars DownTo 0 Begin

MaxH=Highest(MODE_HIGH,shift+PeriodHerst,PeriodHerst-1);
MinL=Lowest(MODE_LOW,shift+PeriodHerst,PeriodHerst-1);
Average=iMA(PeriodA,MODE_SMA,shift);
swap=0;
for i=0 to PeriodA-1 
 {
  swap=swap+pow(open[shift+i]-Average,2);
  swap=swap+pow(high[shift+i]-Average,2);
  swap=swap+pow(low[shift+i]-Average,2);
  swap=swap+pow(close[shift+i]-Average,2);
 };
Deviation=sqrt(swap/((PeriodA-1)*PeriodA));  
//for i=100 downto 2 {ArrayOne[i]=ArrayOne[i-1];};
ArrayOne[1]=(High[MaxH]-Low[MinL])/Deviation;
SetIndexValue(shift,ArrayOne[1]);
swap=0;
If shift< CountBars-PeriodA then {for i=1 to PeriodA {swap=swap+GetIndexValue(shift+i);};swap=swap/PeriodA;}
else swap=0;
SetIndexValue2(shift,swap);
If shift>0 Then AccountedBars=AccountedBars-1;
End;




And here is an explanation of the code and txt file.

<br / translate="no"> Dart



Registered: 27.12.2004
Messages: 122

View profile

PostAdded: Thu Feb 03, 2005 6:23 pm Post Title: Normalized Spread Method or Hear_st Reply with quote Edit/delete this post
As far back as two hundred years ago the mathematician Hurst investigated the processes of river flow etc, the random deviation from the mean were normally distributed. Now the theory has been extended and extended to be called
random walk processes with memory.
Well, as far as I can judge from statistical samples, cloze(shift)-cloze(shift+1) distribution is normal, then the price, or rather its graph, is this very "random walk with memory".
In his works, Hirst obtained an interesting regularity, that was later checked with the computer simulation - R/S=(t/2)^H
where R is (max-min) for some period t
S is a standard slope of the value
then R/S is the expected normalized range value at period t
t - time of the interval in question
H - Hurst coefficient, for the given value is constant
Since we do not know H (although it can be found experimentally), but for a constant time interval (t-const), we can assume that R/S will be constant, or to be more exact, distributed around a certain average normal law with sk = (abs((t/2)^H - (t/2)^(H-0.09)) + abs((t/2)^H - (t/2)^(H+0.09)))/2.
Then, supposing the above hypothesis is true, in case R/S becomes larger than some average, the movement that caused R/S growth must bend, like a pullback or vice versa if
If R/S is less than the average, its growth is expected, i.e. the movement.
I myself have conducted some experiments with this shit, but the result is zero. Although there were a couple of nuances which I did not take into account:
A). I was counting from a fixed average for the period, or maybe I should have counted from a moving average;
B). the intervals (Hearst period and the averaging period) are small from the statistical point of view, but my indicator
(May be someone will optimize it?);
C) I haven't considered the statistical distribution R/S regarding the average, if this is to be done, then we need to introduce
weighting function with the parameters mentioned above;

Like, is there any point in continuing? I have lots of ideas, but not enough time...

RS-Herst.mql
Description:
This is an indicator that calculates R/S and its slippery

Download from
File name: RS-Herst.mql
File size: 1.3 KB
Downloaded: 1 time(s)



Uncle



Registered: 11/17/2004
Messages: 41

View profile

PostAdded: Thu Feb 03, 2005 10:08 am Post title: Re: normalized spread method or Dick_sta. reply with quote
Forget it.
The catch is that the H-parameter for quoting processes is a floating value (like the normalized spread itself on the same period). I once posted a fractal dimension meter (H=2-dimension) on VIAC.

The only thing you can try to use H-parameter for is to select instruments having memory. On currencies it's not a good idea, but on shares it's worth trying. But it may as well be possible to use autocorrelations of increments for these purposes.

P.S. Usually they do not write R/S=(t/2)^H, but R/S=C*(t/2)^H, and H is found as an angular coefficient of the straight line log(R/S)=H*log(t/2)+b by OLS



 
Once again there was a failure of the indicator, but before re-initialising, I noticed that it had been running for almost two weeks in probability display mode. So I saved the picture for history.

 
The problem is that the H-parameter for the quoting processes is floating (like the normalized spread itself on the same period).

If it were constant, it would be unsuitable for indicators.
As for the picture, between August 24 and 25 the price seems to keep moving along the RMS line and the probabilities start to fall sharply enough. Is this due to a glitch, or is there a channel realignment going on?
 
<br / translate="no"> As for the picture, between August 24 and 25, it looks as if price continues to follow the RMS line and the probabilities start to drop quite sharply. Is this due to a glitch, or is there a channel realignment going on?


It is hard to say, perhaps, but the Alert about the indicator failure came just today. Usually after this kind of failure the indicator stops working.
The visible SCO channel is the last drawn by the indicator at 11:00 MSK. Before that there were other channels.
 
Lied (didn't look), for 7:00 server time.
 
On Solandr's advice I am posting some food for thought:

http://forum.traders.kiev.ua/viewtopic.php?t=1310&highlight=&sid=9070cd1888daae94e2bf0605cbd63cb4
http://forum.forex.ua/viewtopic.php?p=14906&highlight=

And the book that was mentioned (the archive is split into two parts):

http://forex.kbpauk.ru/download.php?Number=56226
http://forex.kbpauk.ru/download.php?Number=56228

I highly recommend reading the links, it's quite interesting.
And in general there is quite a lot of information on the subject of interest on White Collar, but finding it is a problem.
 
Posted the EURUSD 30 mins, from 2001 - what I am using now in the tester.
https://c.mql5.com/mql4/forum/2006/08/EURUSD30.zip
I need to change zip extension to rar (according to solandr's recipe). Prepared from the minutes, minutes seem to be close to those posted by HIDDEN ( "Testing for the contest" ), because collected, as I remember, similarly, only from viac.ru I went straight to the spider and took the history to the middle of 2004 there. The volumes should not be trusted, when they were missing they were added arbitrarily, it's just that the tester doesn't work with zero volumes.
 
Posted 30 mins of EURUSD, from 2001 - what I am using now in the tester.


I tested it on this history. The result of working of the Expert Advisor during the period 01.01.2001-15.10.2004 is the loss of -60%.
Testing results of the EA for the period, in which the EA was tested in the post "solandr 15.08.06 07:21" is shown in the image below.

The following conclusions can be made based on the obtained results:
1. The Expert Advisor currently available is "noise-dependent", i.e., the Expert Advisor shows considerable difference in the results obtained from various sources both in terms of the final profit and in terms of performed deals.
2. You can adjust (fit to history) not only parameters of the system but also the trading algorithm, which can probably exist in this EA as well. Parameters of the only confirming oscillator were chosen at the very beginning 3 months ago, just based on visual picture and logic of opening of intraday trades, and they have never been changed since then. All the achievements were made only by modifying the trading algorithm.

Probably, the following assumptions may be made about the reasons that have led to such results. The quotes from different sources may differ within the range of 3-10 pips on the bars of M30 period plus the fact that different brokers have different quotes history holes. Further, when we calculate the Hurst Ratio and wait until its value passes the 0.5 threshold in one or another direction to make a decision concerning a trend reversal, then in the area being close to this value, this difference of 5-10 pips on the bar may play a significant role. Of course, when the Hurst coefficient is far from this area, for example equal to 0.8, there is no principal difference if it is 0.81 or 0.79 (for different quotes). Therefore, since this parameter is one of the key ones for a position opening, its noise fluctuations in the decision making area can affect the result in the same way as it happens with all other indicators. Vladislav in one of his posts, which cannot be found in this thread ;o), wrote that he tested his EA on different brokers and there were cases when one broker potentially profitable position was wiped by stop loss, while the other broker (-s) kept this position in play.

Also I have the assumption that because my realization of the algorithm is certainly different from Vladislav's algorithm, then perhaps the fact of using it for optimization based on the channel potential energy minimax may somehow increase the strategy robustness against noise (differences in quotes from different brokers).

So far I am just observing how the Expert Advisor works on the real account. During 12 trading days the Expert Advisor has made 5 deals. All that time he closed with profit 8.7%. Since such a small number of deals is absolutely statistically insignificant, there is no sense in discussing this result. Let's continue this observation.

In general, I have plans to change this EA into semi-automatic mode. That is, to equip it with possibility to manually open/close a position using the above method of convergence points of the sum of gradients to zero followed by automatic position control using the existing trading algorithm as it can be more profitable according to my observations. Again, we must check and observe. And between observations let's read the book the link to which was kindly shared by Tier. Maybe some new ideas will appear thanks to it?

 
Representing a chart as a polyline and recognizing the image (shape) of the polyline as a whole and its individual fragments.

Is this approach to automated trading promising and feasible?
1...127128129130131132133134135136137138139140141...309
New comment