Machine learning in trading: theory, models, practice and algo-trading - page 2072
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
Nice work! You did a great job...
But what's the use, I did not use neural networks, I searched for the most similar part of the history, the repeatability is 50/50, of course there are series when it coincides. And if we attach more time to the patterns (like prices) the repeatability on the history is practically nil. This leads me to my question - is the movement of 100 points of the price during 20 minutes or 60 minutes the same or not?
Also, it seems that if you use in the search for a similar section not only the time series line but also two lines with variants of events, then the accuracy is higher.
From this follows the question - is the movement of 100 pips in 20 minutes or 60 minutes the same or not?
The speed of the formation of the interval of ZZ greatly affects the decision-making in my models.
How do you determine the long and short shoulder?
If it is more than the previous one in points, then it is long.
If it is more in pips than the previous one, then it is long.I see, thank you. So for the strategy I should take into account this length for stops...
I see, thank you. So for the strategy we should consider this length for stops...
I took a zigzag with a fixed step. Condition: a new knee has been formed, the forecast = 'long' , we put a stop after the last extremum and take it at the point of the previous extremum. (maximal if to buy, minimal if to sell.)
What's the use?
Well, I told you what could be wrong ;)
I did not use a neural network searched for the most similar part of the story repeatability 50/50
If you had used a neural network you wouldn't have found anything at all, it would have smoothed the whole thing out, and you gained experience and understanding!
So the question follows - is the movement of 100 points of the price during 20 minutes or 60 minutes the same thing or not?
Everything is relative, we should first compare the pattern with the volatility.
Also, like, if you use in the search for a similar area not only the time series line but also two lines with the variants of events, the accuracy is higher.
Well, you just wrote above that if you add a condition (time) then the pattern does not find any. The axiom: the more conditions in a pattern, the less of them on the history!
In machine learning it's called "the curse of dimensionality" Wiki helps
10000 iterations. If we assume that expectation=0, then the maximum deviations are +-18000. Your result is 28280, it seems to be a real pattern.
Under the worst scenario the system will start to give profit after 2300 trades.
If you just calculate and take 4 sko as a threshold, then for 7345 trades 4 sko = 16120.
Ugh, how angry - I identified indicators from the standard delivery, which give a different indicator at one point in time, if they were run in different periods in the tester - I do not know how they are counted, but for machine learning they are dangerous!
This also includes iAD() - perhaps these types of indicators use the accumulation from all of history, but as you know, in the tester there is no all of history. As an option, modify the algorithm to clean the indicator once a quarter, then you can learn. It's interesting that the model feeds on those indicators, it likes something about them...
I took a zigzag with a fixed step. Condition: a new knee formed, forecast = 'long', stop after the last extremum, take at the point of the previous extremum with a similar extremum. (maximal if to buy, minimal if to sell.)
Why would the take point at the extremum, because the correction might have been 50%, which means that we should take a stop at around 100% from the previous interval.
10000 iterations. If we assume that expectation=0, then the maximum deviations are +-18000. Your result is 28280, it seems to be a real pattern.
In the worst scenario the system will start to make profit after 2300 trades.
If you just calculate and take 4 sko as a threshold, then for 7345 trades 4 sko = 16120.
Interesting calculations. So you can develop the strategy, the main thing is to understand how to teach it...