entrance point - page 7

New comment
 
fate писал(а) >>

Depending on the degree of point closeness you may, for example, get a coefficient that would help more. When I was testing by hand, on a strong trend (H4) out of 20 EAs 6 EAs with -6+6(bars) showed entry points and vice versa, not even 2 coincided on the same trend and there's nothing to speculate about, I checked and was convinced of that

There are different methods of expert evaluations. They differ in complexity. For example, the simplest one - the sum of all experts' signals is taken and if buy signals prevail, we buy. There is a fuzzy logic method - the output is obtained through a special convolution of all experts' signals where each signal is assigned a weight (there is such an Expert Advisor in the codbase). There are methods of combining Expert Advisors using NS - associative machines (well described by Haykin). There are many methods in general, read references.

 
Mathemat >> :

It's still elementary, but completely wrong maths.

If we assume that signals of three Expert Advisors are independent, then the required probability is 1 - (1-0.55)(1-0.65)(1-0.75) = 1 - 0.03975 = 0.960625.

I'll add my five cents, albeit late.

Expert Advisor's probability is the ratio of profitable trades to all trades.

That is, for the first one, the ratio of profitable trades to unprofitable ones is 0.55 / (1 - 0.55) = 1.2222, for the second one - 0.65 / (1 - 0.65) = 1.8571, for the third one - 0.75 / (1 - 0.75) = 3.0000.

Multiplying these ratios, we get - 1.2222 * 1.8571 * 3.0000 = 6.8095 - the final ratio of profitable trades to non-profitable ones.

If the output is y = x / (1 - x), the inverse is now x = y / (1 + y).

From this we deduce the probability of "united" EA - 6.8095 / (1 + 6.8095) = 0.8720. That is, the final ratio of profitable trades to all ones is 0.8720. It is also the Expert Advisor's probability.

P.S. Why multiply (1.2222 * 1.8571 * 3.0000), I do not know! I even wasted an hour and a half trying to check it programmatically. It turns out 0.8720 (on average).

 
FION >> :

There are different methods of expert assessments. They vary in complexity. For example, the simplest one - the sum of all experts' signals is taken, if buy signals prevail - we buy. There is a method of fuzzy logic - the output is formed by convolution of experts' signals and each signal is assigned a weight (there is such an Expert Advisor in the codbase). There are methods of combining Expert Advisors using NS - associative machines (well described by Haykin). There are many methods in general, read references.

With methods all clear that it not one it is certainly necessary and with it to understand but me the technical question for this purpose and this theme has started I have already written
I'd be very grateful if someone would tell me the best way to combine them all into one EA or connect to one chart, that's all I need.


 

Regarding simple summation, as an example. I took someone else's turntable, pulled out a piece on determining the strength of the trend, minimized the code (the original one consumes too much memory).

In the upper right corner it displays up/down strength in percents. If it's more than 75% - you're in. Here http://highwayfx.ucoz.ru/forum/3-8-1 at the end of the page there is also a version with history.

Files:
trendforce.mq4  8 kb
 
Figar0 >> :

For example, one advisor is classical, the other according to the stars, the third according to folk omens). And using different TA strategies on the same data will not work. And how to count what is a big question.

You could, for example, use the MAI... allows to combine experts of completely different nature...

Neutron >> :

As a first approximation, we can assume that p increases almost linearly with the number of indicators (see the above figure). In its turn, the probability of getting n indicators to work simultaneously decreases exponentially with the growth of the number of indicators, and it means that the frequency of performing deals will decrease as quickly. Thus, we have two competing processes: the profitability and the frequency of trades.

- You cannot assume that p increases linearly with the number of indicators used... more like a parabola :)


- Who says that all indicators should work simultaneously ... Just the optimal number of indicators (on the parabola it will be -b/2a)


But, the conclusion is still disappointing - the less indicators to use in trading - the better. The most optimal is one!


- Combine several indicators into one and you will be happy!


 
fate писал(а) >>

I know it's not just one method, but I'm interested in the technical issue.
I would be very grateful if someone would tell me how to combine them into one EA or connect to one chart.

Here's a simple variant

double Signal(int i){
       double val1 = iWPR(NULL,0,14, i);
       double val2 = iDeMarker(NULL,0,14, i);
       double val3 = iRSI(NULL,0,14,PRICE_CLOSE, i);
       double BBL = iBands(NULL,0,20,1,0,PRICE_CLOSE,MODE_LOWER, i);
       double BBU = iBands(NULL,0,20,1,0,PRICE_CLOSE,MODE_UPPER, i);
      // 
// ------------ buy       
       int BufferUP = 0;
       if( val1>=(-20))  BufferUP ++;
       if( val2>=0.7)    BufferUP ++;
       if( val3>=70)     BufferUP ++;
       if( BBU<=Close[ i]) BufferUP ++;
      
//--------------sell       
       int BufferDN = 0;
       if( val1<=(-80))  BufferDN ++;
       if( val2<=0.3)    BufferDN ++;
       if( val3<=30)     BufferDN ++;
       if( BBL>=Close[ i]) BufferDN ++;
       
       
       return( BufferUP- BufferDN);
}      
 

got the story back

blue - UP
red =Down
white = school difference between UP and Down percentages

Files:
trendforce_2.mq4  9 kb
 
Vinsent_Vega >> :

- You can't assume that p increases linearly with the number of indicators used... More like a parabola :)

Where does the parabola come from, Vincent? And how does it behave at values close to zero (the probability cannot be negative)?

 
Neutron >> :

Hi Alexey.

It's easier to play the Monte Carlo way.

You're really something, Sergey. I have yet to comprehend your conclusions, thank you for the food.

But here's a question for you: why doesn't the probability increase with p=0.5?

 
Mathemat >> :

Where does the parabola come from, Vincent? And how does it behave at values close to zero (the probability can't be negative)?

don't take it so seriously, Alexey... I'm just approximately formulating my practical observations - if the number of experts (indicators) grows, probability of a correct forecast will go up for some time (though not necessarily)... and then begins to fall... I.e. there is a certain optimal number of them, which should be used...

123456789101112
New comment