Machine learning in trading: theory, models, practice and algo-trading - page 1560
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 have always said that machine learning is not suitable for trading.
Forex should be approached like in war games, where the situation is changing all the time and you have to act according to the situation, using special tactics and strategy.
In Forex, it is enough to consider the trend and support/support levels. I think that's what Gerchik is always talking about. But it's not enough.
The outcome is affected by many "insignificant" factors, such as traction, acceleration, rate of price change, decay, reversal and several temporal factors, as well as such a notion as defining a mini-trend on a large trend, which seems to be crawling along the big trend, since dynamically it is impossible to determine the beginning and the end of the trend in any way.
Who has understood, has understood, and who hasn't, won't understand.
I think you wrote something about api economic calendar. I would like to know if it works and develops well, because this topic was somehow stalled on the forum. I have some doubts, whether it is worth to look into it (after the disappointment of statistical library).
It doesn't work in the tester. In online still have bugs.
What if we try to teach the grid to give a probabilistic forecast...? But not the way it is usually done - the signal level corresponds to the probability of prediction, but this is how:
We divide the probabilities from 0 to 100% into, say, 10 sections, which would correspond to 0, 10, 20... 90, 100%, then teach the grid so that for example at 10% signal level should come out 10 correct answers out of 100 in similar cases and so on for each of the sections. Count each area, calculate the standard deviation for each area, and, accordingly, ff get a simple and clear - reducing the standard deviation on average for all probability areas. Of course, in this case, the requirement of statistical validity leads to an increase in the sample for training, but it will be possible to directly use exactly the probability model, for example enter. when the probability prediction exceeds 70%. The thought is pure probabilistic prediction training without futile interpreting of output neuron level into probability.
SZZ the breakdown into sections is done for simplicity, without the need to interpolate the results during training and reduce the requirements in computing power, although it may be necessary to interpolate.
I think you wrote something about api economic calendar. I would like to know if it works and develops well, because this topic was somehow stalled on the forum. I have some doubts, whether it is worth to look into it (after a disappointment from statistical library).
In the tester does not work. There are still bugs in the online.
I don't have it in the tester, after that I lost interest at once, what's the point then. In mt4 it was possible to draw information in the tester at least through webreaests. At present time webreavers and websockets do not work in the tester. You can do it in python, use a database like quandl.com
Inaccessibility in the tester is sad, but it can still be bypassed by some crutches - you can try to read the news from a pre-prepared file, for example. Bugs online - something much more serious and if the MCs will not work on them, then there is no point in being hooked on this api. On the example of the statistical library, we can see that the lack of interest among the general trader masses leads to a loss of interest in the MC as well. Therefore, the lack of interest in the forum on this topic is alarming.
I want to estimate the degree of news influence on prices using the Bayesian methods for myself.
PS. I read a little English forum about MK calendar - I was not optimistic, rather the opposite.What if we try to teach the grid to give a probabilistic forecast...? But not the way it is usually done - the signal level corresponds to the probability of prediction, but this is how:
We divide the probabilities from 0 to 100% into, say, 10 sections, which would correspond to 0, 10, 20... 90, 100%, then teach the grid so that for example at 10% signal level should come out 10 correct answers out of 100 in similar cases and so on for each of the sections. Count each area, calculate the standard deviation for each area, and, accordingly, ff get a simple and clear - reducing the standard deviation on average for all probability areas. Of course, in this case, the requirement of statistical validity leads to an increase in the sample for training, but it will be possible to directly use exactly the probability model, for example enter. when the probability prediction exceeds 70%. The thought is pure probabilistic prediction training without futile interpreting of output neuron level into probability.
Breakdown into sections is done for simplicity, without the need to interpolate the results in training and reduce the requirements in computing power, although it may be necessary to interpolate.
However, I have to try it again, maybe I was wrong somewhere.
But so far my opinion is that there are no regularities in the price behavior and that the price is a SB.
To effectively use the MO apparatus, you can't use a "bare" price series.
It is necessary to take into account what the SB does not have - temporal patterns.
The best "works" is the dependence of price dynamics on the day of the week, worse - daily fluctuations in activity
Inaccessibility in the tester is sad, but it can still be bypassed by some crutches - you can try to read the news from a pre-prepared file, for example. Bugs online - something much more serious and if the MCs will not work on them, then there is no point to be tied to this api. On the example of the statistical library, we can see that the lack of interest among the general trader masses leads to a loss of interest in the MC as well. Therefore, the lack of interest in the forum on this topic is alarming.
I want to estimate the degree of news influence on prices using Bayesian methods for myself.
PS. I read a little English forum about MC calendar - I am not optimistic, rather the opposite.The entire infrastructure is a formidable swamp, if based on trading, but not on the market.
The whole infrastructure is an impassable swamp if you rely on trading and not on the market
Yes, it seems that everything is set up in such a way that traders are built into the mainstream (indicators, grids, martingales, ...) or give it all up.
"Don't show off, Comrade Ivanov! Listen to your song "Valenki"!