Machine learning in trading: theory, models, practice and algo-trading - page 3595
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
Well, everyone goes crazy in their own way here)
The approach closest to me is to construct probability distributions and look for their deviations from what they would be if they were randomly wandering.
It is necessary that they pay royalties to the author)
The official science is precisely that macroeconomics is very bad at predicting exchange rates.
Imho, from a practical point of view, if macroeconomics improves the forecast even a little, it is worth using it in addition to prices. This is quite consistent with the idea of ensemble in the IO.
It seems to me that official science may be wrong or hiding something from the public.
I can show an example of using macroeconomic data (at) an indicator. This is not an advert for a product. It is not and never will be in the marketplace.
I think this method is used by certain groups of people with a lot of money.
You can evaluate for yourself whether it is possible to make money on such data. The indicator does not redraw.
The white-green bar shows the probability. NS without MO. Simple and convenient.
Now you can throw tomatoes.
I don't quite understand how it looks like in your case. In mine I get label distributions for each cluster, similar to what another Alexey did with quantum cuts :) but only I do almost zero work to analyse them independently.
I think it was already partially discussed in the thread when they talked about probabilistic learning (then everything was quickly reduced to the task of classification and calibration of probabilities). To quote myself)
"A broader approach is interesting, with models for which the output is not a number, but any (within reasonable limits, of course) distribution. An example would be the MO used in reliability theory (where the distribution of lifetime is studied) or in probabilistic weather forecasting (where a probability distribution is constructed for the possible amount of rainfall, for example)."
For trading, for example - let the stop loss be fixed in every trade. Take profit is not fixed, instead we know the maximum price advancement in the desired direction before the stop is triggered. We consider this maximum advancement as a random value. Our MO algorithm should find the dependence of the distribution of this random value on the attributes (the output is not a number but a distribution). We compare the distribution with what it would have been under SB and enter if there are differences. There is also a possibility to calculate take profit for each trade separately (according to a specific type of differences in distribution from the SB-variant at given values of the trait), instead of one for all.
I think it has already been partially discussed in the thread when they were talking about probabilistic learning (then everything was quickly reduced to the problem of classification and calibration of probabilities). To quote myself)
"A broader approach is interesting, with models for which the output is not a number, but any (within reasonable limits, of course) distribution. An example would be the MO used in reliability theory (where the distribution of lifetime is studied) or in probabilistic weather forecasting (where a probability distribution is constructed for the possible amount of precipitation, for example)."
For trading, for example - let the stop loss be fixed in every trade. Take profit is not fixed, instead we know the maximum price advancement in the desired direction before the stop is triggered. We consider this maximum advancement as a random value. Our MO algorithm should find the dependence of the distribution of this random value on the attributes (the output is not a number but a distribution). We compare the distribution with what it would have been under SB and enter if there are differences. It is also possible to calculate take profit for each trade separately (according to a particular type of differences in distribution from the SB-variant at the given values of the trait), instead of one for all.
.
.
.
That's cool, did you pick up the key?)
Maybe someone is interested in daily fluctuations of EURUSD spread this year. The nature of small peaks at time XX:00 (marked with grey vertical lines) is interesting.
On gold, for example, such peaks are much more weakly represented.
Perhaps it is time to add round price levels to round time levels)