The neural network only opens trades at night. Did she discover scalping?

 
I trained them several times. Each of them has a different profitability schedule, but each of them opens trades exclusively at night and in the early morning.

The feeling is that I decided to trade Asian strategies.

The sample is small, just started, but the behavior of the network is curious, despite the fact that the most "monetary" periods are the Americas and Europe.

Does it mean that the neural network has found a working strategy? Or is it better to train it so that it trades all day long?



 
how many transactions during the training period?
 
Is this the work of your Neural Machine? I like the fact that there is only one currency pair and only one lot size. Question 1: How many orders are open at most at one time? Question 2: Is it possible for the Neural Network to work with only one open order, only one pair and only with a fixed lot?
 
Andrey Dik:
how many trades on a training period?

Different runs have different numbers. Took the first one I could find:

Training: 25 January 2018 to 18 April 2018 (3 months),
Test period: 01.01.2017 to 31.12.2018 (2 years),
Forward: January 1, 2019 to present (3 and a half months).

Result:

During the training period: 51 trades



During the testing period: 387 trades




On forward: 43


 
Vasily Belozerov:
Is this the work of your Neural Machine? I like the fact that there is only one currency pair and only one lot size. Question 1: How many orders may be opened at most at one time? Question 2: Is it possible for the Neural Network to work with only one open order, only one pair and only a fixed lot?

Yes.

1. A maximum of 20 trades.

2. It is possible, of course.


In the long term, you can load the rest of the majors as well.

There is an interesting function in the programme: you can input data from other currency pairs, and the program will look for patterns in their joint behaviour. Thus, you can simultaneously feed all currencies, metals, indices, stocks, everything in the terminal. As long as the computer does not explode.

Since it's all a pain in the ass, I didn't get to it. Later I'll try to train the eur on data from EURUSD and GBPUSD, or gold and oil, or all the major currencies that have a dollar basis. Let's see what happens later.

A little later I will open a signal and a pamm of another advanced variant of the neural network from the userAndrey Dik. I have a great picture there too:


training - 10 months, test on period after training - 6 months.
The picture shows a small period.


But the result on the real is curious: all trades are opened in the morning so far, although no intraday timeframes were set in the training. It is as if the market is looking for calm.
 
Ivan Butko:
I trained them several times. Each of them has different profitability schedule, but each of them opens deals exclusively at night and early morning.

I feel like I've decided to trade Asian strategies

The sample is small, just started, but the behavior of the network is curious, despite the fact that the most "monetary" periods are the Americas and Europe.

Does it mean that the neural network has found a working strategy? Or is it better to train it so that it trades the whole day?



Can it choose which hours to trade?
 
Ivan Butko:
I trained them several times. Each of them has different profitability schedule, but each of them opens deals exclusively at night and early morning.

I feel like I've decided to trade Asian strategies

The sample is small, just started, but the behavior of the network is curious, despite the fact that the most "monetary" periods are the Americas and Europe.

Does it mean that the neural network has found a working strategy? Or is it better to train it so that it can trade all the day long?



Does your neural network account for spreads?
Maybe it trades at night because it has a nice graph. But the huge spread expansions at night will not allow it to make money in the real world.
 
multiplicator:
Can it choose which hours to trade?
That's what I'm asking, can it?) was curious about this behaviour myself. Not a single trade to america or europe

Yes, it takes into account the spread. Look at the attached screenshot in the thread, it's real 4 days.
 
Ivan Butko:

Yes.

1. A maximum of 20 transactions.

2. You can, of course.


In the long term, you can charge the other majors as well.

There is an interesting feature in the programme: you can input data from other currency pairs, and the software will look for patterns in their combined behaviour. Thus, you can simultaneously feed all currencies, metals, indices, stocks, everything in the terminal. As long as the computer does not explode.

Since it's all a pain in the ass, I didn't get to it. Later I'll try to train the eur on data from EURUSD and GBPUSD, or gold and oil, or all the major currencies that have a dollar basis. Let's see what happens later.

A little later I will open a signal and a pamm of another advanced variant of the neural network from the userAndrey Dik. I have a great picture there too:


training - 10 months, test on period after training - 6 months. (white line).


But the result on the real account is quite curious: all transactions are opened in the morning hours, although no intraday periods were used in training. It's as if the market is looking for calm.

Is it possible to see somewhere a chart of balance and TIME for one pair, with one order in the market, with a fixed lot?

 
The way she was trained, that's how she works.
 
Ivan Butko:
That's what I'm asking, can it not?) was curious about this behaviour myself. Not a single transaction to america and europe

Yes, it takes into account the spread. Look at the attached screenshot in the thread, it's real 4 days.
You have a purchased black box.
That's the disadvantage of it, you can't modify it yourself.
See its documentation, maybe there is such a possibility.
You have no one to ask)) We don't know the device, official pre-sale description or documents issued to buyers.