Machine learning in trading: theory, models, practice and algo-trading - page 3440

[fxsaber]

Maxim Dmitrievsky #:

Doing it in my toaster, no problem with that. The problem is how to make it work when there is a spread :)

On what symbols?

[fxsaber]

Forester #:
У вас есть какая то статистика, сколько лимиток остаются не исполненными в среднем?

Strongly depends on the time of day, the broker and the symbol itself.

Last week on one of the brokers.

Rejects = 14.29% (6/42, 00:00:00.047), Slip = +327.47%, 100%
Rejects = 400.00% (76/19, 00:00:00.426), Slip = +217.49%, 100%
Rejects = 55.56% (15/27, 00:00:00.194), Slip(sim) = +371.94%, 100%
Rejects = 141.67% (34/24, 00:00:00.283), Slip = +74.81%, 100%
Rejects = 50.00% (2/4, 01:23:45.796), Slip = +90.17%, 94.4% = 3.40/3.60(3)
Rejects = 49.47% (47/95, 00:00:02.117), Slip(sim) = +341.84%, 100%
Rejects = 35.71% (5/14, 00:00:00.294), Slip(sim) = +95.54%, 100%
Rejects = 44.15% (181/410, 00:00:21.685), Slip(sim) = +51.73%, 100%
Rejects = 7.34% (8/109, 00:00:00.103), Slip(sim) = +83.01%, 100%
Rejects = 20.51% (32/156, 00:00:03.662), Slip(sim) = +67.45%, 100%
Rejects = 54.47% (128/235, 00:00:13.478), Slip(sim) = +130.03%, 100%
Rejects = 9.48% (11/116, 00:00:00.401), Slip(sim) = +145.41%, 100%
Rejects = 182.35% (31/17, 00:00:03.906), Slip = +128.20%, 100%
Rejects = 48.33% (29/60, 00:00:03.821), Slip = +116.50%, 100%

Each line is a symbol.

How much time should we give for execution? It seems to me that after 1-2 minutes you should take them off, as the situation will change.

I have a different approach. I have a real-time virtual machine with perfect execution (tester), and the combat Expert Advisor just tries to get the same execution levels as on the virtual machine. Thus a limiter can get even 10 redirects, but then it will be executed. Or it can be cancelled after one re-jacket. Much depends on the situation in the virtual machine.

[[Deleted]]

fxsaber #:

On what symbols?

any

[fxsaber]

Maxim Dmitrievsky #:

on any

EURUSD with zero spread (minus news and rollover) is a standard of the modern industry. You can only fight over the size of the commission: 5-10 per million turnover - you can find it if you want.

[[Deleted]]

fxsaber #:

EURUSD with zero spread (minus news and rollover) is the standard of the modern industry. You can only fight over the size of the commission: 5-10 per million turnover - you can find it if you want.

The commission may not be beaten then... I am talking about ways of filtering trades + MO, so that he trades in places where signals potentially exceed markups.

I guess you could put it this way. That is, you need to set an inflated markup during training so that the algorithm goes OOS with it.

I asked an orthodox AI to write me such a code, after which it broke even the one that worked.

[mytarmailS]

we and saber live in different realities...

In trading - we try to predict, he exploits inefficiencies.

In optimisation - we are trying to find a working TS through optimisation and avoid retraining, and he has a working TS and just wants to find optimal settings.

etc...

That's why we will always speak the same words but about different things, and we will never understand each other.

[[Deleted]]

mytarmailS #:

Toaster is honest, compared and checked with commercial the whole book of transactions, not just by eye on the capital curve.

It's exactly the same with the opens, I checked.

[mytarmailS]

Maxim Dmitrievsky #:

It's exactly the same thing with the Oppenheimers, I checked.

On the minute?

[[Deleted]]

mytarmailS #:
On the minute?

including

[mytarmailS]

Maxim Dmitrievsky #:

including

Evra, 1 minute, blue is the difference between cloze[i] and open[i+1], red is the cumulative sum of blue.

How can you have a normal and also coincide with the tester I in my heart not ...