Machine learning in trading: theory, models, practice and algo-trading - page 446
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
I got 14 parameters for optimization. I am afraid that this number will result in overfitting.
http://radikal.ru/video/Eq6HXmqO3mH
Beautiful.
I'm trying to switch from classification to regression, but so far it hasn't worked out to give up classes completely. Using price increment as a target, training a model, predicting something, it's not a problem. I even get r^2 >0, but so far no more than 0.1. The problem is that funds charts built when trading with this model look bad - with big drawdowns, long periods in minus. There is no such uniform money growth as when optimizing model on recovery factor of trade.
That's why I rounded prediction to classes -1 and 1 and plotted them and used something like recovery factor for fitness function.
So the question is, what do you use for the fitness function, just r^2, or some offshoot of r^2 that makes the errors evenly distributed over time?
I'm working on the first point.
My TP/SL is somehow complicated and unclear. If I take almost any working robot and start optimizing its take and stop, then it starts to lose profit using new data. Such optimization leads to overfitting to specific pips and drawdowns of the past that will never happen again, while the robot will be waiting for them.But if before optimizing the robot some constant values of TP and SL are set for it and its other parameters are optimized, then with a good chance everything will turn out well.
Very strange, just change the order of TP/SL optimization at the beginning or at the end and the result will be quite different.
Yes, we can reduce the harm, but why? The big question is, what does tp/sl even give us, in addition to the quasi-optimal portfolio that is lined up with AI predictions???
I'm not afraid to claim absolutely nothing. If the AI is good, the ideal strategy is simply to rebalance the portfolio based on the probabilities of future asset gains/declines. If, for example, your AI says that an asset will grow, but it has gone down on the noise beyond the SL level, should you close the position? I do not think that is wise. In the case of advanced algorithmic trading, the classic t/sl does not make sense. But it is reasonable and even necessary to have a system SL, that is, algorithms that recognize when a strategy or AI for some reason decreased efficiency or even went astray, then you should stop trading with it.
. In the case of advanced algorithmic trading, classical tp/sl does not make sense. However, it is reasonable and even necessary to have a system SL, that is, algorithms that recognize when a strategy or AI for some reason decreased efficiency or even went astray, then you should stop trading on them.
Totally agree with you, as I wrote above, maybe not quite clear.
SL is an alarm signal that the TS is not working, its risk has exceeded everything seen on the history. We need to fix the loss and start dealing with the trading system itself.
SL is an alarm signal that the TS is not working, its risk has exceeded everything that was seen on the history. It is necessary to fix the loss and start dealing with the trading system itself.
Exactly. Not necessarily the TC, the alarms may be very different.
Lack of margin, for example.)
That's why I rounded prediction to classes -1 and 1 and plotted means and then used something like recovery factor for fitness function. so question, what do you use for fitness function, just r^2, or some branch of r^2 that makes errors evenly distributed over time?
Doc, this is intimate))) yes different. The fact that it doesn't work as it should only indicates that the model is unsuccessful. Subsequent
use of anything, is nothing but creating an ensemble of models(although the target during optimization may be different)...
For example, the lack of margin.)
Somehow the SL was triggered when the Internet went down. SL is an emergency termination, usually it does not come to the triggering.
Somehow the SL was triggered when the Internet went down. SL is a general emergency termination, usually it does not come to the triggering.
It takes part in trailing and just as a part of the system, at some levels... How do you do it without the sl, trades are closing anyway. You may close not by market price but pull the sl, it's always safer that way, especially against connection failure.