Has anyone made Automatic Virtual Self-Optimization for their robot? - page 6

New comment
 
fxsaber:

Surges are not as bad as their frequency. That is, you need to earn more on the quiet stretch than you lose on the surge. And to do that, you need the quiet period to be relatively long. The ability to squeeze a profit out of the quiet part is to a large extent the merit of self-optimization.

Quiet segments are often absent. Or they are quiet on a different time scale.

fxsaber, tell me from experience what is the optimum historical period for calculation in bars/days?

Thank you!

 
Vitaly Muzichenko:

fxsaber, tell me from experience what is the optimum historical period to calculate in bars/days?

I don't know and I don't think there can be a clear answer. I primarily focus on the number of non-overlapping trades per symbol. It would be desirable to have at least a hundred of them in a month. But these are the TPs I am interested in. I.e. active, and thus heavy.

For example, classical nocturnals, of which I have seen decent ones in Signals, trade an hour or two a day and up to three trades. That is, it is very little. The authors usually optimize them over a couple of years, making several hundred trades during that period and practicing filtering not only by time. Most often they do it with several pairs in the hope (not without reason) that they will draw each other one by one.

The risk is lower, so jumps between quiet periods do not affect them much. And there is time to think about it. I.e. there are many advantages of such an approach, if we ignore the sluggishness of action.


On the interval. Here today did the optimization for the last two weeks and the best pass applied to a few two weeks before that (on each character - multitester). Did the same for three weeks. In principle, this should be enough to define a quiet period. They have one constant property - almost any adequate TS earns on them without over-optimisation. And there is one more property - no matter how one tries to show off with the logic of TS, there will be a definite unbreakable profit ceiling on this period. That is, I have never managed to create TC logic, which is noticeably better than others.


Therefore sometimes there is a necessity (desire) to opt not for a quiet period, but for a jump-like one. Like, let us not lose where it is bad, because where it is good, it will earn anyway. But there is a nuance here in the form of a bounce. I.e. on the next bounces will be losing anyway.


As a result, for stability, you will again slip back to sluggish overnights, which you do not want to do.

 
Vitaly Muzichenko:

Thank you! Where in the EA should this be entered?

if(ProfitToday() >= 100500%) 
   ForexRemove();

P.S. Maybe try it this way :)

and on the bogamas

 
fxsaber:

If the TS is unable to adjust to a consistently negative spread, then self-optimisation should help - the TS should become profitable.

If self-optimisation when the spread is negative does not work, there is probably something wrong with the TS.

With such correct TS the same problem constantly arises during research. Space profits on the real history of some short sections. These profits are not systematic, because the negative spread is veiled there. As a result, you optimize such an Expert Advisor and see a perfect result. When you look, you see that half of the profit for the month is calculated within several hours.

I have to cunningly arrange filters for super profitable parts of the history to avoid this. I am inclined to believe that these are broken quotes, because it is unlikely that lag arbitrage has taken place for real profits.

 
fxsaber:

With such correct TCs, the same problem keeps cropping up during research. Cosmic profits on the real history of some short sections. These profits are not systematic, because the negative spread is veiled there. As a result, you optimize such an Expert Advisor and see a perfect result. When you look, you see that half of the profit for the month is calculated within several hours.

I have to cunningly arrange filters for super profitable parts of the history to avoid this. I believe that these are broken quotes, because it is unlikely that lag arbitrage has taken place for the real profit.

And what do negative spreads have to do with it. I don't have a lot of them, and besides, there are brokers who never have negative spreads. Also, I just skip them, although on the real account I see negative spreads very rarely,

They are not that scary.

 
Yuriy Zaytsev:

and on the bogamas

It's boring out there.

 
fxsaber:

With such correct TCs, the same problem keeps cropping up during research. Cosmic profits on the real history of some short sections. These profits are not systematic, because the negative spread is veiled there. As a result, you optimize such an Expert Advisor and see a perfect result. When you look, you see that half of the profit for the month is calculated within several hours.

I have to cunningly arrange filters for super profitable parts of the history to avoid this. I'm inclined that these are broken quotes, because it's unlikely that lag arbitrage has taken place for real profits.

cut out chunks like that

and in cast. symbols in the tester, do the ticks go without lag?

It is written that in real they are generated once in 100 ms. I.e. they cannot even be used for realtime arbitrage

 
Petros Shatakhtsyan:

What do negative spreads have to do with it?

There are veiled negative spreads - lag arbitrage, where the negative spread is spread out over time: not one-sentence.

 
Maxim Dmitrievsky:

cut out pieces like this.

and in cast. characters in the tester, do the ticks go on without a delay?

It says that in real life they are generated every 100 ms. I.e., they cannot even be used for realtime arbitrage

The formula custom ticks - there are 10 Hz. Your custom ones are as you prescribe. But it has nothing to do with the Tester.

And in the Tester the ticks go exactly the same way as for other symbols.


On such periods you can simply disable trading in the Tester.

 
fxsaber:

The formulaic custom tics are 10Hz there. Your custom ones are as you prescribe. But this has nothing to do with the Tester.

In the Tester, the ticks are exactly the same as for other symbols.


On such periods, you can simply disable trading in the Tester.

Ah, right, thank you. I.e. the formula ones are meant

There is one funny algorithm for arbitrage in one brokerage house, I may write an article later

1234567891011
New comment
Reason: