Machine learning in trading: theory, models, practice and algo-trading - page 1574

 
Boris:

Colleagues, why don't you stop measuring with rulers and tell simple peasants how to check synthetics for non-training?


This is the third time I've asked, but it's still there

Forward test

 
Dmitry:

The forward test

What should be its duration if the average order is open for 5-10 months?


A year?

Совершение сделок - Торговые операции - Справка по MetaTrader 5
Совершение сделок - Торговые операции - Справка по MetaTrader 5
  • www.metatrader5.com
Торговая деятельность в платформе связана с формированием и отсылкой рыночных и отложенных ордеров для исполнения брокером, а также с управлением текущими позициями путем их модификации или закрытия. Платформа позволяет удобно просматривать торговую историю на счете, настраивать оповещения о событиях на рынке и многое другое. Открытие позиций...
 
Boris:
What should be its duration if the average order is open for 5-10 months?

At least 10 months.

 
Boris:

What should be its duration if the average order is open for 5-10 months?


A year?

Why a year? The model cannot work that long - it is easier to re-optimize it regularly.

 
Dmitry:

Why a year? The model can't work that long - it's easier to re-optimize it regularly.


Well, if the model has been built on analysis for 13 years without any reoptimization, how can it be reoptimized? and most importantly - why?

to do this you need to change the basic principles of operation

 
Boris:


well if the model was created on the analysis of 13 years without any reoptimization, then how can it be reoptimized? and most importantly - why?

you need to change the basic principles of work

I don't get it.

You created this model by fitting its parameters to the training sample - that's optimization.

You can find out if the model is overoptimized or not by a forward test. If the model on the training sample does NOT CRITICALLY WASTE its values, then, Bombay, there is no overoptimization

 
Dimitri:

I don't get it.

You created this model by fitting its parameters to the training sample - that's optimization.

You can find out if the model is overoptimized or not by a forward test. If the model on the training sample does NOT CRITICALLY worsen its values, then, Bombay, there is no overoptimization

Okay, let's smoke for a year.
 

Gentlemen, don't use the words: push, write, and write in the same sentence

or you'll end up with a lot of butter.

Keep it short, please.

 
Dmitry:

At least 10 months.

Forward from one deal?

 
Andrey Khatimlianskii:

A forward of a single trade?

forward is not deals, but the quality of synthetic or its probabilistic characteristics

Reason: