Some signs of the right TCs - page 28

[fxsaber]

Aleksey Mavrin:

Which TS are we talking about and more importantly which instrument?

Scalper, cross.

[Алексей Тарабанов]

fxsaber:

Scalper, cross.

Expected. Applicable?

[fxsaber]

Алексей Тарабанов:

Expected. Applicable?

Yes.

[fxsaber]

"Generally correct" TS - the best optimization result for all parameters is invariant to the voiced transformations.

It is possible to substitute this EA function in the test EA.

// https://www.mql5.com/ru/blogs/post/734146
input group "EA"
input bool inEven = false;
input datetime t_enter = D'01.03.2019';
input datetime t_exit = D'01.12.2019';

void EA()
{
  MqlTick Tick;

  if (SymbolInfoTick(_Symbol, Tick) && (Tick.time >= t_enter))
  {
    if (Tick.time >= t_exit)
    {
      if (OrderSelect(0, SELECT_BY_POS))    
        OrderClose(OrderTicket(), OrderLots(), OrderClosePrice(), 0);
    }    
    else if (!OrderSelect(0, SELECT_BY_POS))
      OrderSend(_Symbol, inEven, 1, inEven ? Tick.bid : Tick.ask, 0, 0, 0);  
  }
}

To make sure that this TS is "generally correct".

Intermediate results:

• Mathematically correct TS - any pass is invariant to tsVR transformations.
• Generally correct TS - the best/worst pass of Optimization in all parameters is invariant to tsBP transformations. Optimization criterion - relative profitability.

[Aleksey Nikolayev]

fxsaber:

"Generally correct" TC - the best optimization result in all parameters is invariant to the sounded transformations.

Intermediate result:

• Generalized Correct TS - best/worst Optimization pass in all parameters is invariant to the transformations of the TSV. The criterion of Optimization is relative profitability.

It is quite usual for mathematics to have a constructive definition in addition to the formal one. The first one is convenient for speculative conclusions, and the second one for solving applied problems. And, unfortunately, they are not always equivalent and one has either to put up with this or try to take it into account in calculations - there is a lot of it in the matstat.

The only thing I want to clarify in your version is that optimization does not necessarily apply to all possible parameter sets. Your variant of my "Expert Advisor" shows well that entry and exit moments must be excluded from optimization - otherwise the actual transformation of quotes will turn out to be absolutely different from the studied one. Another thing is that for a meaningful EA the exact definition of the required subset of parameter sets is hardly possible.

[Aleksey Nikolayev]

fxsaber:

I'll answer the same here about the lack of custom character capabilities in MT (you wrote that blog communication is not quite convenient).

I may be wrong, but if we need to test on a large enough number of custom characters, the only option is to make one very long custom character out of them. If we need optimization on a sufficiently large number of custom characters, then this method doesn't seem to be suitable anymore.

I can vaguely see some possibility to apply "generalized correctness" in building a portfolio of one EA with different parameters. For this purpose we conduct optimization on the set of transformed quotes. The transformation of quotes consists, for example, in cutting them into a given number of pieces and then rearranging and gluing them together in all possible orders.

[fxsaber]

Aleksey Nikolayev:

According to your version of my "Expert Advisor" it is well visible that entry and entry moments should be excluded from optimization - otherwise the actual transformation of quotes will turn out to be quite different from the one being studied.\

If Even is involved in the optimisation, it is possible to optimise the rest of the parameters.

[fxsaber]

Aleksey Nikolayev:

I vaguely see some possibility of applying "generalised correctness" in constructing a portfolio of one EA with different parameters. To do this we perform optimization on a set of transformed quotes. The transformation of quotes consists, for example, in cutting them into a given number of pieces and then rearranging and gluing them together in all possible orders.

A portfolio of different sets is done differently. As for the custom characters, I did the mixing of daily intervals. Perhaps for some TC it makes sense. But not in my case.

Optimize a lot of custom symbols - MultiTester.

[Dmitry Fedoseev]

The only thing left to do is to choose one profitable system from this plethora of systems - and that is a difficult task. The only thing that remains is to choose one profitable system out of all this abundance - and that is a difficult, impossible task.

[Dmitry Fedoseev]

Nikolai Semko:
I would add the main thing:

• has no period-dependent parameters.
• operation of the TS does not depend on the current chart timeframe
• the operation of TS does not depend on the symbol-instrument
• the whole setting of TS - is only the setting of risk management (the size of the used deposit)

Utopian fantasy