Machine learning in trading: theory, models, practice and algo-trading - page 113
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 tried a similar approach with my data. I have more than two models in my committee. I could not get a unanimous answer from them, but if you trade when at least 80% of the models agree with the answer - you get better trading results, I advise you to try.
questionable....
how much better ?
Now there is a selection of model parameters and predictors, I took intermediate results from there. The committee itself is still weak, we need to further adjust the parameters, prediction accuracy is about 55%, it is not enough to overcome the spread. I have added the requirement of 90% goodness of agreement (third chart) - profit has gone up.
Charts - profit in pips eurusd, trade on d1 for the last six months.
Now there is a selection of model parameters and predictors, I took intermediate results from there. The committee itself is still weak, we need to further adjust the parameters, prediction accuracy is about 55%, it is not enough to overcome the spread. I have added the requirement of 90% goodness of agreement (third chart) - profit has gone up.
Charts - profit in pips eurusd, trade on d1 for the last six months.
This is nice...
i wonder if it makes sense to do so with RF or it's the same as adding trees to the same model
I have added
How do you make these committees?
questionable....
how much better?
For us it is not known and here's why.
The fact is that Michael is not at all of our type: he has no evidence about the future. In his case, the network is part of the TS, and the decisions from the NS turn out to be interconnected with other elements of the TS. So it is not possible to isolate the effectiveness of the NS proper, and it is not interesting for him.
This TS is usual for TA, but with bows and trinkets and it does not solve the main problem: complete uncertainty of the future. Yes, we optimize every bar/day/week and everything is fine. Once we fully trust the TS, then the next drawdown will not be a drawdown, but will turn out to be a drain.
For us it is not known, and here's why.
The fact is that Mikhail is not at all from our line
You still can not understand the difference between the classifier and predictor. For example, this morning I trained separately buy, separately sell..... The result is for today, it is quite possible that I was lucky, but how :-)
Although, maybe your systems are related to the forecasters, I do not know.......
beautiful...
I wonder if it makes sense to do this with RF or it's the same as just adding trees to the same model
You still can't understand the difference between classifier and forecaster. For example, this morning I trained separately buy, separately sell..... The result is for today, it may be so lucky, but it is so :-)
Although, maybe your systems are related to the forecasters, I do not know.......
You still haven't written your article on how classification differs from regression and from prediction. Waiting ;)
Also, how are these committees made, are there packages or manually?
You can always try a committee with rf. I don't think the applied package matters, as long as the models in the committee already have some ability to predict new data, otherwise this percentage filter is useless (why listen to the majority if it's not right? :)
You still haven't written your article on how classification differs from regression and from prediction. Waiting ;)