Machine learning in trading: theory, models, practice and algo-trading - page 2533
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
That's why it is better to collect your own history in real time, which will not be changed in the past (which is not uncommon sin brokers) when you trained the network, and it initializes incorrectly in a couple of days, because the broker decided to change something in the history and such practices and evidence abound. And when the history is written in your own files and no one can change it but you, then everything starts to work fine.
Just everyone forgets that in addition to training, there is also initialization of the EA, say, after the weekend, and it is very important that the initialization was correct, on the data on which it was trained, otherwise the subsequent signals will be significantly different from those that were trained by the network ...
That's why it is better to collect your own history in real time, which will not be changed in the past (which is not uncommon sin of brokers) when you have trained the network, and it initializes incorrectly in a couple of days, because the broker decided to change something in the history and such practices and evidence abound. And when the history is written in your own files and no one can change it but you, then everything starts to work fine.
It's just that everyone forgets that in addition to training, there is also initialization of the advisor, say, after the weekend, and it is very important to initialize it correctly, on those data which it was trained on, otherwise the next time signals will be very different from those ones the network was trained with ...
To collect the history yourself is the most reliable. But this can only be done with those DCs to which you are connected at the time of collection.
And if you want to switch to a new one? But this one has different conditions, commission, spreads.
In order to trade with this new brokerage company: the model must see and take into account all this, i.e. we need to study on its data, not on the data of another brokerage company or ideal DukasCopi. In the complaint may not ask why your data is not like in the other brokerage company?
For example you trained on data with a low spread and decided to trade with a brokerage company with a higher spread but no commission. For example on a 5 pt. If it works earlier, a new trade will be executed earlier. Maybe, the price may change and TP will trigger instead of SL (it will not reach the 5 points to that SL). I.e. there will be a completely different trade.
I agree with Maxim that DSOs/changes may add trash to quotes. But it is more practical for them to copy the quotes from the reference ones. So that no one claims for the different candlesticks. This can be seen immediately just by re-logging between brokerage companies. The eye will immediately notice the changes.
The kitchens are more likely to draw the necessary candles in the trading process, and to replace them in the history with the reference ones in a few days.
For myself, I think the best way is to use the data of brokerage companies that are going to trade.
If the tests on their history are much worse than others (the added garbage will worsen the learning curve), then just give it up.
If the trade will be worse (with comparable tests) then draw candles in the process, and also get away from such.
Okay... The story is up to everyone's personal discretion.
But how to take into account ECN account commission in training - this is a common problem. It cannot be added to the spread (4-5 pt) because it will distort the triggering moments.
PS: and swaps too. If not intraday trading.To collect the history yourself is the most reliable. But this can only be done with those DCs to which you are connected at the time of collection.
And if you want to switch to a new one? But this one has different conditions, commission, spreads.
In order to trade with this new brokerage company: the model must see and take into account all this, i.e. we need to study on its data, not on the data of another brokerage company or ideal DukasCopi. In the complaint may not ask why your data is not like in the other brokerage company?
For example you trained on data with a low spread and decided to trade with a brokerage company with a higher spread but no commission. For example on a 5 pt. If it works earlier, a new trade will be executed earlier. Maybe, the price may change and TP will trigger instead of SL (it will not reach the 5 points to that SL). I.e. there will be a completely different trade.
I agree with Maxim that DSOs/changes may add trash to quotes. But it is more practical for them to copy the quotes from the reference ones. So that no one claims for the different candlesticks. This can be seen immediately just by re-logging between brokerage companies. The eye will immediately notice the changes.
The kitchens are more likely to draw the necessary candles in the trading process, and to replace them in the history with the reference ones in a few days.
For myself, I think the best way is to use the data of brokerage companies that are going to trade.
If the tests on their history are much worse than others (the added garbage will worsen the learning curve), then just give it up.
If the trade will be worse (with comparable tests) it means the candlesticks are being drawn in the process, and to get rid of that too.
Nonsense. The problem is solved simply - they write everywhere on their site and insert in the trading conditions that they take quotes "from several liquidity providers and mix them for the benefit of traders. That's all. After that they don't even need to change the history
Nonsense. The problem is solved simply - they write everywhere on their website and insert in the terms of trade that they take quotes "from several liquidity providers and mix them for the benefit of traders. That's all. After that they do not even need to change the history.
Quotes of the liquidity providers and at a better price are EUN accounts. And they are usually very good and almost indistinguishable from the reference ones (which also publish quotes of liquidity providers).
But what if it is a kitchen without ECN? That's where you can run into a drawing candles for internal use and trash in the history. Which then it is desirable to mask with a reference.
PS: There are fewer and fewer kitchens, though. Even Alp... ECN (which can not be called a "kitchen") regularly changes the domain, because the new one gets under blocking almost once a week.
Quotes from liquidity providers and even at a better price are EUN accounts. And they are usually very good and almost indistinguishable from the reference ones (which also publish quotes of liquidity providers).
But what if it is a kitchen without ECN? That's where you can run into a drawing candles for internal use and trash in the history. Which then it is desirable to disguise with a benchmark.
Have you ever read the rules of brokerage companies? All brokerage companies write that they take quotes from several banks and mix them. It is clear that this is a lie, but it justifies any price manipulation and there is no need to correct the history (except for too large outliers).
The only way out of the situation is to build a TS with a large TR. Everything else is a game with pips, which is guaranteed to be won by the broker.
Have you not read the rules of the DC in all these years? All DCs write that they take quotes from several banks and mix them up. It is clear that this is a lie, but it justifies any price manipulation and there is no need to correct the history (except for too large outliers).
The only way out of the situation is to build a TS with a large TR. All the rest - the game with pips, which is guaranteed to win the broker.
I read it in ECN. And when you give them a deal of 1 - 2 lots, the execution increases to a few seconds - apparently and it is true that they take out to the liquidity provider what they cannot ock inside.
Yes, I got the winning model at TP=SL=300 pt. But I did not take into account the commission. IO cannot do it.
I read at ECN. And when you give them a deal for 1 - 2 lots, the execution increases to a few seconds - apparently it is true to the liquidity provider, what is inside they can not ocean.
Yes, I got the winning model at TP=SL=300 pt. But I did not take into account the commission. MO can't do that.
Lena?!
Tell me more about it.
This approach is very interesting.
I don't want to write nonsense with a clever look again, it's a super classic idea, ridge regression was invented in1970.