An accident or an unrecognised pattern? - page 4
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
Colleagues. A thought has occurred to me about a quality forward test, on quotes inheriting all properties of the pair.
Question.
Is it realistic to use the same 10 year quotes as for the main test, but reverse them in time. That is to go backwards?
To start the test on December 31. 2018, and end on December 31. 2008.
Thevolatility and tick intensity in time will remain the same...
Question... how to reverse the quotes or programmatically ask the tester to process the history backwards?
Colleagues. A thought has occurred to me about a quality forward test, on quotes inheriting all properties of the pair.
Question.
Is it realistic to use the same 10-year quotes as for the main test, but reverse them in time. That is to go backwards?
To start the test on December 31. 2018, and end on December 31. 2008.
Thevolatility and tick intensity in time will remain the same...
Question. how to reverse the quotes or programmatically ask the tester to process the history backwards?
This is not a forward
turn them backwards in time. You mean go backwards?
By doing so, you are changing cause and effect relationships (on which you can only make money) to "backwards".
Don't make up nonsense. You can't make a synthetic with inherited quote properties for many reasons.
You've already been written many times about what you should do - divide your 10 years of history into two segments of 5 and 5.
On one of them the development and optimization, on the second (later) the forward testing. The technique is simple and long known in machine learning.
You can divide 5 years of forward section into 5 pieces of one year each, it will show stability better.
By doing so, you are changing cause-effect relationships (on which you can only make money) to "the other way round".
Don't make up nonsense. You can't make a synthetic with inherited quote properties for many reasons.
You've already been written many times about what you should do - divide your 10 years of history into two segments of 5 and 5.
On one of them the development and optimization, on the second (later) the forward testing. The technique is simple and long known in machine learning.
You can divide the 5-year section of the forward one into 5 pieces of one year, it will show the stability better.
It won't show anything.
It's just that in the original form it fitted all 10 years at once, and in other cases it will fit in chunks - same eggs, only in profile
By doing so, you are changing cause-effect relationships (on which you can only make money) to "the other way round".
Don't make up nonsense. You can't make a synthetic with inherited quote properties for many reasons.
You've already been written many times about what you should do - divide your 10 years of history into two segments of 5 and 5.
On one of them the development and optimization, on the second (later) the forward testing. The technique is simple and long known in machine learning.
You can divide the 5 year forward section into 5 pieces of one year each, this will show stability better.
Don't be so categorical.
I agree with you that causality, and other tick sequence type conditions will change dramatically...
But I have an algorithm that works on probabilities, on an eagle/tails type with some conditions) and it is not sensitive to event sequences.
Just like heads and tails are not sensitive to the history of previous outcomes and the sequence in which they are accounted for.
Dimitri below correctly points out that the tests I've been running for 10 years... I have the same algorithm for the whole story.
There is no point in crushing the history. But I crushed it)))
The results are the same.
Not enough quality history or another method of verification.
It's not a forward.
OK, not a forward...
But such a check (forward) would be of quality in your opinion?
OK, not a forward...
But would such a check (reversed) be qualitative in your opinion?
Run it on random rows - if the TS will "detect" the same "patterns" there, then there is most likely nothing in your system.
And if the TS will not profit on random rows - then it exploits a real LAWYER, which is not present on random rows, but which is in the price
OK, not a forward...
But such a check (forward) would be quality in your opinion?
Lots of things can be thought of. For example, one could randomly mix up the real increments and summarize them in this new order.
The only question is, what is the point?
It's just that in the original form he was fitting all 10 years at once, and in other cases he will fit in chunks - same eggs, only in profile
No. Forward is the section where no fitting was done.
Don't be so categorical.
I have an algorithm that works on probabilities, like heads/tails with some conditions) and it is not sensitive to sequences of events.
Just like heads and tails are not sensitive to the history of previous outcomes and the sequence in which they are accounted for.
It lacks qualitative history or another method of verification.
I'm being categorical because I know what I'm talking about. You want to waste years of your life on fallacies, your right)
Any algorithm, with or without probabilities, predicts future price behaviour. By going backwards through a quote, you render your algorithm meaningless.
There are no other methods, except for forward, and there is no need for it. Everything was invented by clever people long ago, you just have to use it. This is not a subject for discussion at all.
If people on this forum read textbooks on machine learning before engaging in discussions, many questions would go away by themselves)