Discussion of article "Naive Bayes classifier for signals of a set of indicators" - page 2
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
Then the topic of another article - how to find indicators that are independent of each other automatically?
And then we almost have a simple KnowHow for creating robots from any indicator collection. Further, you can add trailing, mangement and so on to the basic algorithm via the MQL5 Wizard.
Yes, there is such a topic. It has already been dealt with by real scientists, which I am not - you can find articles on the Internet. I do not have ready material, and it will probably take a lot of time to elaborate it. At the same time, there are already publications on the topic on mql5.com, such as this one or this one.
That said, there are already publications on mql5.com on the topic, like this one or this one.
Thanks for the links and reminder, but that's not it - not a popular mechanic.
Thanks for the links and reminder, but that's not it - not popular mechanics.
Did I get the idea right that the links provided are difficult to understand and practical, and now just require a ready-made "working car" without all the Excel and R stuff? This is me clarifying for future potential authors. ;-)
Did I get the idea right that the references given are difficult to understand and apply in practice, and now you just want a ready-made "working car" without any Excel and R? This is me clarifying for future potential authors. ;-)
Sure - everyone knows how to use smartphones, though no one knows the basics of radio engineering and other complex science that goes into them.
A naive Bayesian classifier requires a strong form of independence from a set of features (in our case, indicators) - independence in the aggregate (not just pairwise or uncorrelated). I have come across the statement that such independence cannot be obtained for a set of ordinary, meaningful indicators.
A naive Bayesian classifier requires a strong form of independence from a set of features (in our case - indicators) - independence in the aggregate (not only pairwise or uncorrelated).
It will be possible to compare theoretical results of calculations with practical results of testing strategies based on several indicators. I think that 3 indicators in one strategy is the ceiling, you can't insert more.
Most likely, you are right. From the theoretical point of view, there can be more independent indicators (even equal to the number of bars on which we count), only they will not be very meaningful. But it is quite possible that we can compose some number of normal indicators (or close to them) from them.
The considerations are approximately as follows. Suppose we have n bars. Let's define the average price p(i) for each of them, for example p(i)=(open(i)+close(i))/2. The set of random variables p(1),...,p(n) will of course be dependent. But it is known that a series of prices is close to be considered as a series of independent increments. Therefore, the set of n random variables d(1)=p(2)-p(1), d(2)=p(3)-p(2), ... d(n-1)=p(n)-p(n-1), p(n) will be close to independent. Now, any set of functions from our set will be independent if any argument is included only in the expression of one of them. To put it simply: the set for four bars from functions I1(d1,d2) and I2(d3,p4) will be independent, but I1(d1,d2,d3) and I2(d3,p4) will not be independent, because of d3.
For example, two different MAs will always be dependent. But if you take two MAs such that the second one is shifted back in time by the period of the first one, then a system of the first MA and their difference will be independent.
I'll try to recount in my own words more simply:
1. Let there are three strategies:
2. All three indicators A, B and C do not correlate with each other - i.e. they give signals to enter the market independently of each other.
3. It is required to calculate the theoretical percentage of winning trades for the Start_ABC strategy, in which market entry occurs only if all three indicators show entry in the same direction simultaneously.
Then P(Win|ABC) = P(Win|A)* P(Win|B)* P(Win|C) /[ P(Win|A)* P(Win|B)* P(Win|C) - (1 - P(Win|A))*(1 - P(Win|B))*(1 - P(Win|C))) ]
Thanks! Then it turns out that P(Win|ABC) is always greater than each stratum individually.
Then the topic of another article is how to find indicators that are independent of each other automatically?
And then we have almost ready a simple KnowHow on how to create robots from any collection of indicators. Next, you can add trailing, mangement and so on to the basic algorithm via the MQL5 Wizard.
Take an arbitrary window (N) and stretch it over time, building a correlation matrix for each moment. Then add up all the matrices and get the average. The closer to zero the corresponding cell value of the average matrix is, the more independent the indicators are between each other over the length of the interval N.
Probably, in R this is done in one line, since, obviously, such a task is one of the first in statistical research.
Thank you! Then it turns out that P(Win|ABC) is always greater than each stratum individually.
only if all strategies are mutually independent and give a probability greater than 0.5.