Advisor for an article. Testing for all comers. - page 6

New comment
 
jelizavettka:
I have noticed that a certain small change to the parameters giving the most successful forward leads to less profit deviation than the same change in other groups of optimized parameters. - If you know how to detect it correctly, you won't need a forward. The parameters giving a successful forward have a bigger stability margin. (IMHO)

Unfortunately, extremes run, i.e. do not stand still. Therefore we have to catch them with forwards. There is no other way of doing it yet. But, if the forward is successful, then the probability of him continuing behind the right side is quite decent due to the safety margin.


Theoretically, during optimization it is possible to randomly create small deviations of input parameters, i.e. deviate them slightly in different directions. It seems it should lead to better results. But we need to test and try it out. Theory does not always coincide with practice, especially in non-stationary conditions.

 
TheXpert:
Forward is needed in any case. Otherwise, how else can we evaluate?

Forward is just for checking assumptions. Even if you take usual muwings. You change parameters (optimized) back and forth by a certain percentage in each line (this is a simplification) - look - one line gives the smallest deviation from the total financial result before changes in the parameters. - then forward this line to the parameters, .... you do other forwards for comparison - and oh my God! - it, that very line, turns out to be the most successful forward as it has a larger margin of stability of the parameter combination to the market.
 
Reshetov:
But if the forward is successful, the likelihood of him continuing behind right-back is quite decent thanks to his margin of safety.
Erm, is there a rationale for margin of safety in the article?
 
Reshetov:
Unfortunately, extremes run, i.e. do not stand still. Therefore we have to catch them with forwards. There is no other way of doing it yet. However, if a forward is successful, then the probability of him continuing behind the right wing is quite decent thanks to the margin of safety.

I completely agree. I think there are councillors who do not give successful forwards at all))
 
jelizavettka:
Forward - just to check the assumption. even to take an ordinary muving. change the parameters (optimized) back and forth by a certain percentage in each line (this is a simplification) - look - one line gives the smallest deviation from the total financial result before changes in the parameters. - then forward this line to the parameters, .... you do other forwards for comparison - and oh my God! - it, that very line, turns out to actually give the most successful forward.
There you go... If only it were that simple. I mean, it's simpler, but not there at all... Ah, well, wait for the article, we'll talk about it.
 
TheXpert:
There you go... I'm just complimenting you (if only it were that simple. I mean, it's simpler, but not there at all... Ah, well, we'll wait for the article, and then we'll talk about it.

The trouble here is that these assumptions are very difficult (for me at least) to verify and test using software methods at MCCUL.
 
TheXpert:
Erm, is there a rationale for the safety margin in the article?

No. But R. Pardo has it. I.e. after optimization and successful forwarding, we select a single input parameter, disable the genetic algorithm and run the optimization over the whole range. We look at the results. Preferably, there should be only one extremum and its edges should be smooth. It is natural that the value of the parameter identified upon full optimization should be located near the top but not necessarily at the very top as the extremum will not stand still.

This is how all input parameters need to be checked for lousiness.

We can also check in pairs but it is inconvenient on MT4 as in the tester extrema are highlighted in dark green and visible only from above in the optimization results (upon pressing the space bar). But on MT5 you can already admire the tops in the 3D graph.

 
jelizavettka:

The problem is that these assumptions are very difficult to check and test using software methods on MMKYL (at least for me).
On MT5, of course, it is easier, because there you only need to set the size of the forward test and run the optimization. The tester displays a bunch of successful and not very successful forwards. All you need to do is to select the best one. The only pity, as compared to MT4, the optimization is very slow there, but it is possible to wait for its completion, if you connect the Clouds Network and spend a few cents.
 
Reshetov: Theoretically, it is possible to randomly create small deviations of input parameters during optimization, i.e. to deviate them slightly in different directions.
The optimisation is going through them anyway, so where else are they to deviate?
 
Mathemat:
The optimisation is already overriding them, so where else can they be deviated from?
It does, but not all combinations during genetic optimization. Therefore it is useful to do it in the selected combination.
12345678910
New comment
Reason: