Machine learning in trading: theory, models, practice and algo-trading - page 445
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'm trying to master a very specific model - GARCH. It attracts me by the fact that the original series is decomposed into components and then these components are modeled separately. And the decomposition into components is intuitively understandable and directly related to retraining of the model. Since I'm interested only in the model's retraining capability (I can create retrained Expert Advisors in TA with lifetime up to 6 months), that's what determined the choice of GARCH.
I am not aware of any approaches in NS that would allow to account for thick tails, kinks... It seems to me that the NS model itself has nothing at all to do with the problems of the original quotient.
In GARCH, while fitting the model, I can run tests that serve as a basis that in the future the resulting model will behave exactly as it does on the training data. At the moment I don't know how to fit GARCH, the parameters of which would be significant with more than 90% probability.
What package do you use? I hope fGarch?
Good luck.
Which package do you use? Hopefully fGarch?
Good luck.
rugarch
Dear users of NS, scaffolding and other methods of MO.
After receiving the forecast, how do you trade?
There are variants:
1) we get a signal, open, wait for the reverse signal - i.e. pre-rotation trading, without TP and SL
2) p. 1 + inversions if the signal is not reversed, but following the previous one
3) point 1 + TP and SL (how to select TP and SL - with the MT optimizer or manually)
4) point 2 + TP and SL
5) Any of the points + trailing, there are many types of trailing, which one? (How do you select the trailing parameters - with the MT optimizer or manually).
Maybe some other variants?
Dear users of NS, scaffolding and other MO methods.
After receiving the forecast, how do you trade?
There are variants:
1) we got a signal, we open, and wait for the reverse signal - i.e. we conduct pre-trading without TP and SL.
2) p. 1 + shares if the signal is not reversed, but the tail signal
3) point 1 + TP and SL (how to select TP and SL - with the MT optimizer or manually)
4) point 2 + TP and SL
5) Any of the points + trailing, there are many types of trailing, which one? (How do you select the trailing parameters - with the MT optimizer or manually).
What are the other options?
I should warn you right away that I don't have a working system on MO yet). But the ideology remains the same, similar to my systems on logic.
1. No predictions are made in principle. I.e., where and what will really go - no assumptions. The signal is received when the situation in the quotes statistically suits this or that deal. And, of course, it can be erroneous. Yes, and in my systems the idea is that MO should supplement, not replace, the system based on logic.
2. Further there is the usual support of the deal. For simplicity, we can consider trailing. There is no TP and SL. There is a protective stop for emergencies such as disconnection of the Internet and other surprises.
Other comrades may do things differently. I'm not aware of that.
Dear users of NS, scaffolding and other methods of MO.
After receiving a forecast, how do you trade?
There are variants:
1) received a signal, opened, wait for the reverse signal - i.e. pre-rotation trading, without TP and SL
2) point 1 + shares, if the signal is not reversed, but the previous one
3) point 1 + TP and SL, (how do you select TP and SL - with MT optimizer or manually)
4) point 2 + TP and SL
5) any of points + trailing, there are many types of trailing, which one is it? (How do you choose the trailing parameters - with the MT optimizer or manually).
Maybe some other variants?
I use the forecast as an indicator for how much to keep in my portfolio and in what direction. I would not trade manually, it would be useless and even harmful for robots.
Dear users of NS, scaffolds and other MO methods.
How do you trade after you get the forecast?
TP\CL - for manual trading, for robots it makes no sense, even harmful.
However, if before the optimization of the robot some constant TP and SL values are set and its other parameters are optimized, then in the end everything will probably turn out well.
It's very strange, just change the order of whether to optimize TP/SL at the beginning or at the end and the result is completely different.
I use SL, fixed, yes through the optimizer. In case of 15 trades it is of course retuning in the tester, and if I have 200+ trades and stop is optimized within 25 to 40 points, then I believe it is not retuning but simply searching for the best value. I have not specified takeprofits and close orders by a common trawl + opposite signal, also through the optimizer in a narrow range. I cannot do without stops, if only portfoliowise, but it's the same without them. We can optimize anything and it will not be a superfitting if the number of solutions for the TS is limited and a decent testing period is specified. If TS is not limited in variants of solutions, then of course it will be superfitting. At increase of optimization period signals are smoothed and as though noisy signals are eliminated by themselves, but here still depends on NS (number of nerons), if very little then classification will be too rough and system will be low-profile, if a lot then there can be overfitting, though it already depends more on predictors instead of NS, if space of solutions (variants of predictor combinations) is limited then there will be no overfitting.
My TP/SL is somehow complicated and incomprehensible. If I take almost any working robot and start optimizing its take and stop, 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.However, if before the optimization of the robot some constant TP and SL values are set and its other parameters are optimized, then in the end everything will probably turn out well.
It's very strange, just change the order of whether to optimize TP/SL at the beginning or at the end and the result is completely different.
Entering and exiting a position - that should be handled by the algorithm for making the appropriate decision, such as a forest, NS, or just a bunch of indicators.
The SL should not have anything to do with the problem of the position at all. The SL is not a trading system parameter and should not be optimized.
The SL is a risk management. With absolutely any system of decision making on a position, a decision on the maximum allowable drawdown on the balance is made. It is a matter of psychological comfort, the status of used money (the last, not sorry to lose...). The figure is set, and that's it. The figure of 30% is used in signals. Where did it come from? From nowhere. Did someone optimize it? If you set algorithmic SL = 30% on drawdown in your EA, it will be 70% protection of your money.
Entering and exiting a position - that should be handled by the algorithm for making the appropriate decision, such as a forest, NS, or just a bunch of indicators.
The SL should not have anything to do with the problem of the position at all. The SL is not a trading system parameter and should not be optimized.
The SL is a risk management. With absolutely any system of decision making on a position, a decision on the maximum allowable drawdown on the balance is made. It is a matter of psychological comfort, the status of used money (the last, not sorry to lose...). The figure is set, and that's it. The figure of 30% is used in signals. Where did it come from? From nowhere. Did someone optimize it? If you set the algorithmic SL = 30% on drawdown in your EA, it will be 70% protection of your money.
So you deliberately limit the space of variants in TS, SL is part of the system, as well as there folding or averaging. It results in a great imbalance when the system may get one huge SL and stop, while you may allow many small losing trades and take out due to expected payoffs but not due to the absence of stops.
Any system has an error that is limited to stops and nothing else.
There is no psychology here and should not be, there are system characteristics reproducible on the forward, and tight tolerances when the system stops working with a minimum loss and not 30%. 30% received - this is goodbye. The 10% drawdown is the maximum I think you can work with, if it's more then there is a point of no return in most cases. Moreover, if we are talking about a system that needs to be retrained periodically, that is, you can not predict its stability over a long period of time.
If you take almost any working robot, and start optimizing the take and stop, the robot starts to frantically drain on new data.
It simply adjusts for volatility (between trades) on the sampling. If it changes in the future, then of course the cp will p... If we make a model with forecasts and take it into account during optimization
But we don't need it, rollover with reinvestment is our all))) And the speaker Fomenko will tell us about dispersion, heteroscedasticity, etc.))
http://radikal.ru/video/Eq6HXmqO3mH