Discussion of article "Optimal approach to the development and analysis of trading systems" - page 5

New comment
 
Inquiring:
You strangely ignored the letter from 20.09.2020 to personal mail . You could have at least just two words - your opinion.

I do not remember such a thing at all. You can write in a PM. I don't check my mail at all. Only if something important is sent and I know that they will send what is most important.

 
"I wish it had been in your material..." actually it was, just in the section on the mathematics of optimal search. It's just that I deliberately avoided this kind of analysis for one very important reason. Your function that you want to analyse is a function that maps certain system metrics to your labour, (labour-time is the equivalent). In other words, by introducing this function you claim that by spending a certain amount of time you will get a certain result, in fact you will not always get it, you have only the probability of getting it. There are never any guarantees in Forex and all we can operate with are probabilities. Such analysis as yours is possible only if you fully understand your system, its physics and how it works. It is impossible to determine the type of function itself.) To determine it, you need statistical data from a huge number of developers, and each developer will have a different type of this dependence, and you will have to average these data, or move away from such functions in favour of probability functions, which in my opinion is the best solution.
 
Denis Kirichenko:

Eugene, you seem to be a techie... so, on the subject of optimisation... a couple of thoughts of my own...

The most visual thing is a graph. We see curves, we see dependencies, we see areas, what and where we need to optimise... It's a pity that this was not present in your material....

But of course, to do this, we need to get a view of the curve that describes the law of relationships that we are interested in...

My rough estimate of the resource costs of finding and refining a profitable strategy: most likely we are dealing with a logarithmic law.




On the X-axis are all labour costs and on the Y-axis are the returns of the strategy.


The question then is. At what point should we stop [x,y]? I'd be interested in the opinion of interested developers...

I agree with the logarithmic law, but I would shift the curve on the y-axis to the left, very much to the left). Because the profit entry threshold is not from 0 labour (financial) costs, it is strongly above 0.
 
Maxim Romanov:
I agree with the logarithmic law, but I would shift the curve on the Y axis to the left, very much to the left). Because the threshold for entering profit is not from 0 labour (financial) costs, it is strongly above 0.
He has it right actually, you may have just looked at it inattentively. His profit entry threshold is about X=1.2. Anything less is in the negative part of the values, anything more is in the positive part. Of course, what X is and how it is calculated is another question, and what Y is too, but the general shape of the curve is correct. He has labour costs on the X axis, no questions asked.
 

All figures are arbitrary. The main thing is to convey the meaning. Yes, the blue curve (a(x)) would be called "input-output". In the economic sense, it turns out to be cost-effectiveness.

By the way, I don't agree that the first derivative is useless.... It just shows the rate of change of the dependence. I would think about the point where it is equal to or less than 1.0. The point is that each subsequent unit of expenditure on the trading system should bring at least one unit of income.

 
Denis Kirichenko:

All figures are arbitrary. The main thing is to convey the meaning. Yes, the blue curve (a(x)) would be called "input-output". In the economic sense, it is the cost-effectiveness of costs.

By the way, I don't agree that the first derivative is useless.... It just shows the rate of change of the dependence. I would think about the point where it is equal to or less than 1.0. The point is that a unit of costs for a trading system should bring at least a unit of income.

Hm, well, you have specified, now it turns out that X and Y are almost the same dimension ). Then you need to use a completely different logic. Then the logic of saving time is not applicable to this dependence, but only the logic of optimal use of your resource (money or your labour). then we should do the following:

1) solve the equation a'(x) = 1 , find the root X0. This root is the point at which further expenditures are not profitable.

2) find a(X0 )/x0 . If a (X0)/x0 > 1 then the costs of the system are quite justified. Otherwise, it makes no sense to improve it.

3) If condition 2 is fulfilled, we can evaluate this indicator more deeply. In fact, it will be an analogue of the profit-factor, only in the context of your task.

I will only specify that this is true only for a logarithmic function, if the type of function is different, you will have to adjust the conditions (just the type of logarithmic function gives you the opportunity to use the point a'(x) = 1, ). In other cases, you will have to look for the maximum of A(x)= a(x)/x . Here everything is as it should be. The first derivative, the search for extrema and their analysis, again there should be an interval [X1,X2] because your labour or money costs are limited, and there is no sense in analysing infinity. It can be even simpler. we turn the combined index A(x) into x(A) and analyse this function, everything will be simpler here. As an argument will be the required A>1, the window of which is quite simple to set . Find the first derivative of A and look for minima of x.

 

Beautiful, vital, original story-telling "how to invent a bicycle" with photos of hand-made parts. Bravo, author! )


Only such pictures should not be shown, beginners may believe in them:


 

I like it. The approach is not mine, but it has a right to life. And I think the code in the article is superfluous. It would have been better to elaborate on the approach to testing and selection of Expert Advisors, and you could take any Expert Advisors with the necessary algorithms. And you can compare many conditions, few conditions, significant conditions, not significant.

Normal article)

 
Andrey Khatimlianskii:

Beautiful, vital, original story-telling "how to invent a bicycle" with photos of hand-made parts. Bravo, author! )


Only such pictures should not be shown, beginners may believe in them:


I understand your position. There will be an article on how to do such a thing. It's not appropriate for this article

 
Valeriy Yastremskiy:

I like it. The approach is not mine, but it has a right to life. And I think the code in the article is superfluous. It would have been better to elaborate on the approach to testing and selection of Expert Advisors, and you could take any Expert Advisors with the necessary algorithms. And you can compare many conditions, few conditions, significant conditions, not significant.

Normal article)

I would do as you say but the code is mandatory here. It's required. But I didn't bother with it. The next article will have useful code

1234567
New comment