It's impossible to make money on Forox!!! - page 31
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
Shit, I don't know the statistical terminology, sorry. What statistical indicators of BP change over time? These indicators should be implemented in the form of adjustable parameters in the generator of a synthetic trading tool. This is necessary to detect which statistical parameters of BP will kill the TS.
There is no universal answer to what changes over time. Dependencies in a very broad sense. And changing of BP distribution parameters like mo, dispersion and higher order moments is clear to what it will lead to.
The whole problem is that the synthetic tool by its characteristics, movement features and other nuances may not coincide with a real tool in the future and in this connection all this research work will be done in vain....
The idea is not to create a synthetic VR similar to the real one.
There is no universal answer to the question of what changes over time. Dependencies in a very broad sense. And changes in BP distribution parameters like mo, variance and higher order moments are already clear what will lead to.
I am not looking for a universal answer.
Let's say we created a synthetic instrument 01.01.2009-01.06.2009, M5. We know that the instrument's dispersion (or another statistic parameter) varies throughout the range by the linear law (or any other law that can be chosen in the generator parameters) from such a value to such a value (we set a parameter in the generator of SVR). We ran our Expert Advisor on this tool. We have understood the conditions under which it may work. We corrected the logic. We tested it, etc. We have got a solid Expert Advisor (not a grail, otherwise it will be killed). We have implemented it on the virtual machine. Put it on the real one. Dismissed. They gave up on everything and went to the factory as a turner and miller. :)
Generally, my main goal is to find out the limits of the learning ability of systems with AI. On an artificial tool, it is easier to control the learning process and control where learning would be useless.
An idea has been born. As someone who is not versed in statistics, I will not be able to implement it. But there are such people in this thread (knowledgeable).
The gist. Many traders are of the opinion that the market changes over time. How - does not matter, but it is a problem for TS, solvable or not is another question.
We can create a generator of a synthetic trading tool. It allows adjusting the change in time of the normal/normal distribution of differences and other statistical parameters of BP. Then, testing on such a synthetic trading tool, we can identify weaknesses of TS and improve them to maximize their survivability on changing BP. And so on.
What do you think?
The purpose is clear when you change the parameters of the BP task to immediately see the response, but I'll confront Mischek from such BP use will be little well you pose the TS which is well defines the change in a synthetic BP and then what, how then to move to a real market, would not it be easier to immediately build the TS for the real market.
Where is the guarantee that the patterns you set in synthetic BP are present in the real world?
The idea is not to create a synthetic BP similar to the real one.
I'm not looking for a universal answer.
Let's say we created a synthetic tool 01.01.2009-01.06.2009, M5. We know that variance (or another statistic) of that symbol changes on the full interval by a linear law (or any other law that can be chosen in the generator parameters) from such a value to such a value (we set a parameter in the generator of SVR). We ran our Expert Advisor on this tool. We have understood the conditions under which it may work. We corrected the logic. We tested it, etc. We have got a solid Expert Advisor (not a grail, otherwise it will be killed). We have implemented it on the virtual machine. Put it on the real one. Dismissed. They gave up on everything and went to the factory as a turner and miller. :)
Generally, my main goal is to find out the limits of AI's learning ability. On an artificial tool, it is easier to control the process of learning and control where it will be useless to learn.
Then it's probably better not to synthesize VR from scratch, but take a real + noise generated as you wrote.
The idea is not to create a synthetic BP similar to the real thing.
I'm not looking for a universal answer.
Let's say they've created a synthetic tool ...
...to the factory as a turner-miller. :)
So maybe don't quit the factory :o)
I did almost the same as you, when I was looking for the best extrapolator settings, I found out that with those settings on perfect synthetic harmonics the extrapolator predicts exactly by 100 bar. I found out and calmed down because there are no perfect harmonics on the market. So what you suggest can be done, but just to find out take it as a given and calm down :o)
Of course, it's not worth a penny for the research of all sorts of McDi-like systems. I am well aware of that.
But for researching adaptive TS and TS with AI, it is the best. It is like in the laboratory: you turn the knobs and see the response of the TS.
And if we take real BP + generated noise, how do we know where our adaptive TS stumbles?
Then it's probably better not to synthesise BP from scratch, but to take the real + noise generated as you wrote.
Hmmm... That's an interesting idea. You could even just add a BS. I'll have to think about it.
===
That's what I like about these discussions? One gave an idea - does not roll, the idea itself is impractical, and someone says something like in passing, and already thinking in a new direction. Although we have already gone through the separation of static residues, but here the opposite approach turns out. And the tasks are different.
>> Yeah. I'll have to think about it.)
Of course, it's not worth a penny for researching any kind of mcdi-like system. I am well aware of that.
But for researching adaptive TS and TS with AI, it's just right. It is like in the laboratory: you turn the knobs and see the response of the TS.
And if we take real BP + generated noise, how do we know where our adaptive TS stumbles?
Roger. You want to test fully adaptive systems. If you generate BP from a distribution that changes periodically, then the success of the adaptive systems will depend on how often that distribution changes. Set that the distribution changes at every bar and nothing adapts to it. But if, for example, after 1000 bars, it's fine. If the frequency changes randomly, then the success of the adaptation will depend on the distribution of that randomness.
But of course this doesn't guarantee in any way that the adaptability to the real series will be similar.
got it. You want to test fully adaptive systems. If you generate BPs based on a distribution that changes periodically, then the success of the adaptive systems will depend on how often that distribution changes. Set that the distribution changes on every bar and nothing adapts to it. But if, for example, every thousand counts, then it's fine. If the frequency changes randomly, then the success of the adaptation will depend on the distribution of that randomness.
Yes, you got that right, to test fully adaptive systems. And to be able, on top of that, to change the frequency of change in the distribution.
PS! The harsh realities of real BPs (excuse the tautology), such as gaps, no adaptive system can foresee. But it doesn't need to, because these will be single wrong decisions of the adaptive TS.
Yes, you got that right, to check fully adaptive systems. And to be able to change the frequency of the distribution as well.
Preprocessing, i.e. what is fed to the system as input, is likely to be important here. IMHO, this is a cornerstone of adaptive systems. The values themselves should characterise the stable phases of the market. And synthetics should be generated based on these inputs. Roughly speaking, they should be generated and their distribution should be changed (changing values of input parameters of the adaptive system)