Machine learning in trading: theory, models, practice and algo-trading - page 3343
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
The spread is not a mistake
It is a problem of formulating the target: there is no spread in increments between ascii and increments between bid.
What kind of error can be attributed to the spread? This seemingly stupid problem does not give me a rest. If a model works well with a low spread, but when the spread is increased, it is worse. What exactly to fix?:) or how to train it to trade with a large spread.
.
Use real spread for markup and in signs. Not sure about the signs, not sure it will help, it also turns out that on 3-5 signs the OOS is better than if the model 100-1000 uses. I.e. the chance that the spread will fall into these 3-5 is small.
Maybe it will add a couple of % to our side from almost 50/50, which we get now. Or maybe, as always, it will just take time for all this, without significant result.
I'm switching from bars to ticks, there the spread is real, not minimum per bar. For markup I will use Virtual tester from fxsaber-a (it is possible to make trades and get their results without spoiling the statistics of MQ tester, where I trade only on the signals of the model), then train the model on Saturdays and trade until the next Saturday. It will be a jacking-forward, just like in real trading. But you have your own virtual tester in Python, maybe you don't need it.
The main bread and caviar lies at the spread level, in a large number of short transactions. The logic of finding a balance between the number of transactions and predictability is not quite clear. Increasing the duration of transactions and decreasing their number smoothly leads to loss of alpha.
Mum, send me some woollen socks. We have an electrically heated floor in our cell, but no one knows the code to switch it on).
Use real spread for markup and in signs. I'm not sure about the signs, it also turns out that on 3-5 signs the OOS is better than if the model uses 100-1000. I.e. the chance that the spread will fall into these 3-5 is small.
Maybe it will add a couple of % to our side from almost 50/50, which we get now. Or maybe, as always, it will just take time for all this, without significant result.
I'm switching from bars to ticks, there the spread is real, not minimum per bar. For markup I will use Virtual tester from fxsaber-a (it is possible to make trades and get their results without spoiling the statistics of MQ tester, where I trade only on the signals of the model), then train the model on Saturdays and trade until the next Saturday. It will be a valking-forward, as in real trading. But you have your own virtual tester in Python, maybe you don't need it.
When you increase the spread in the tester, some trades seem to fly off into the minus. But some of them remain. It is possible to train without it at all, but the question is how to run it on the necessary spread and fix the TS, discarding bad trades. I would like to either retrain or add a filter. I can't make up my mind.
Because attempts to build the spread into the training itself do not work.When the spread increases in the tester, some trades seem to fly off into the negative. But some of them remain. It is possible to train without it at all, but the question is how to then run it at the required spread and fix the TS, discarding bad trades. I would like to either retrain or add a filter. I can't make up my mind.
Because attempts to build the spread into the training itself do not work.When mum sends woollen socks, you'll find the code in them. Until then, I'll stay out of your ego's way.
When Mum sends woollen socks, you'll find the code in them. Until then, I'll stay out of your ego's way.
What are you, a complete wreck? Get out of here.
Are you out of your mind? Get out of here.
I can tell by the lingo that the socks haven't arrived yet).
When the spread increases in the tester, some trades seem to fly off into the negative. But some of them remain. It is possible to train without it at all, but the question is how to then run it at the required spread and fix the TS, discarding bad trades. I would like to either retrain or add a filter. I can't make up my mind.
Because attempts to build the spread into the training itself do not work.Probably classically filter, if( Spred > 10 pt ){don't trade and don't mark up}. Or not in pips, average spread * 2 or *3.... *10.
Probably classically filter, if( Spred > 10 pt ){don't trade or mark up}. Or not in pips, average spread * 2 or *3.... *10.
Smiled at the almost certainty of an incredibly clever thought on an incredibly clever thought.
And is everyone around here stupid or what?????
Probably classically filter, if( Spred > 10 pt ){don't trade or mark up}. Or not in pips, average spread * 2 or *3.... *10.