Machine learning in trading: theory, models, practice and algo-trading - page 3547
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
You're bidding the real price. It's gonna make the rules float, too.
No.
Price is perfectly correlated with the model and if it floats by absolute values, it doesn't matter...
If you build a trading rule by stochastic, nobody cares that stochastic has its own range from 100 to -100 (or whatever it is) and does not correspond to the real price.
If you build a trading rule by classification, no one c ares that the rule is built on the curve of the model probability and it does not correspond to the real price.
etc...
the main thing is a fair correlation (with trading rules) and not MSE (prices and models).
No, it's not.
The price correlates perfectly with the model, and if the absolute values float, it doesn't matter...
If a trading rule is based on stochastic, nobody cares that stochastic has its own range from 100 to -100 (or whatever it is) and does not correspond to the real price.
If you build a trading rule by classification, no one c ares that the rule is built on the curve of the model probability and it does not correspond to the real price.
etc...
the main thing is honest correlation, not MSE.
Well, I'll show you the difference later. Maybe it's not critical for you.
GLM can work with different distributions in the residuals, while linear assumes a normal distribution.
GLM can work with different distributions in the residuals, while the ruler assumes a normal distribution.
Yes, glm will give more true coefficients, in theory.
You need a multi label model, I'm not sure if there is such a GLM, and if there isn't, then you should try to train 4 models, and if you already train 4 models, then you can train Forrest too...
The point, as I see it, is not in the models, but in the search for new properties of the model outputs
You need a model multi label is not sure that there is such a GLM-ca, and if not, you need to try then 4 models to train, and if you already train 4 models, you can already forrest then to train.
The point, as I see it, is not in the models, but in the search for new properties of the model outputs
There are many ways to think. Imho, the best way is to think in the direction of what and how to calculate. For example, to calculate the "schedule" of price reversal and not to see it. Drawings of scimitars and other things will not help if they do not lead to calculations.
The "timetable" (the attraction of reversal moments to a certain time) is there. And by the way, it was YOU who tried to calculate the schedule in artificial returns on EURUSD and didn't see it - don't blame your ogrihs on me.
There is a "schedule" (attraction of reversal moments to a certain time). And by the way, it was YOU who tried to count the schedule in artificial returns on EURUSD and didn't see it - don't blame your ogrihs on me.
Well, then you should show this schedule to avoid appearing unsubstantiated, to put it mildly.
Well, then it's worth showing that timetable to avoid appearing unsubstantiated, to say the least.
You open tick volumes (i.e. "activity") and see the same thing.
You open EA stats in the market and codebase and see the same thing. The same chart, exactly, with the same peaks and troughs.
are the probabilities of any event. Including reversals.
how many years can we talk about the same thing? we can discuss it again...there will be another 50-100 pages of repetitions about the same old things, but not directly related to machine learning.
PS/ if a certain method contradicts objectively observable things - then there are questions to the method.