An accident or an unrecognised pattern? - 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
But the main part of the history is adjusting to the history or finding patterns.
If the strategy's returns on the forward are preserved, then a pattern is found, if not, then it's just a fitting to the story.
OK, thank you. I'll give it a try. I'm off to get acquainted with the forward adjustments)
Alexey, excellent article in line with the pattern finding task, especially the sober conclusion about insignificant profits, which coincides with my conclusions. Let's look for new directions in this way. Thanks for the link.
Thank you. It would be quite surprising to get a meaningful profit in such a simple and obvious case) Just, the gaps turned out to be suitable for a simple demonstration of the approach.
Alexey. Thank you, I have read it already and am familiar with the results and methods, as well as your previous article with risk estimation.
Especially I am close to your described method of random walk of price, as in my question (post) I mean exactly this feature. random walk.
But to apply your method as a stencil to the decision of the problem, I do not think yet how, and fundamentally and in applied experience, I am not so strongly savvy as you, that reliably and quickly to solve this problem.
Tell me Alexey, if I provide you with an algorithm that I believe creates a 50/50% probability of event guessing, will you evaluate its credibility or unreliability?
My algorithm for finding a price works on the principle of a theorvert, but it ensures repeatability of the outcome in the entire sample of history, as well as in some parts of it.
It looks like this:
The algorithm has only three variables SL, TP and Market Entry Point.
I set a certain range of values for each of these variables to dissolve/average the influence of fitting.
SL from 40 to 70
TP from 40 to 70
Market entry point from 0 to 12.
Total of 12 493 variables.
Test results on the history of 10 years:
Task.
Identify/prove: Is this result purely a fit or is there an algorithm where the probability of random, independent outcomes may be greater than 50/50.
Alexey. Will you do it?
I am skeptical to my results, i suppose they were caused by mistake in code or logical conditions, but since full week i can't find neither one or the other.
Help... And the diamond of your generosity will shine in the setting of my gratitude)
Unfortunately, not ready to take on your task due to the busyness of my own. I can only share some general thoughts in this thread.
You are absolutely right about forward testing at the section that was not used during optimization. In addition, I can suggest to perform such testing at several sites - so as to get not a single number (e.g. profit), but a sample of several sites. This sample can be tested by the matstat methods for the significance of its mean.
Unfortunately, I am not prepared to take on your task due to the busyness of my own. I can only share some general considerations in this thread.
You have correctly written about forward testing at an unused part of the optimization. In addition, I can suggest to perform such testing at several sites - so as to get not a single number (e.g. profit), but a sample of several sites. This sample can be tested by methods of matstat on the significance of its average positivity.
Thanks Alexey, I started to carry out the forwards, but only on the history segments included in the general test for 10 years, which has already been passed.
It is clear that such a forward passes the test with comparable results and apparently makes no sense.
Here it is a 1/2 forward per sample.
And there is no other (qualitative) history. Hence there is a difficulty - what to forward on?
Maybe it is possible to create synthetic quotes with properties similar to those of the tested symbol?
Then we could use them.
Then a question. Can we create such quotes that somehow inherit the properties of quotes of the tested symbol?
I can easily sample the significance of positivity. (I will add all the variance and get the average, and compare it to a normal distribution).
If I understood the logical methodology of how to distinguish random from regular, I could do the calculations.
I can't understand the methodology yet.
Thanks Alexey, I have started to do forwards, but only on the sections of history that made it into the overall 10 year test, which has already been previously passed.
It is clear that such a forward passes the test with comparable results and apparently makes no sense.
Here it is a 1/2 forward per sample.
And there is no other (qualitative) history. Hence there is a difficulty - what to forward on?
Maybe it is possible to create synthetic quotes with properties similar to those of the tested symbol?
Then we could use them.
Then a question. Can we create such quotes that somehow inherit the properties of quotes of the tested symbol?
I can easily sample the significance of positivity. (I will add all the variance and get the average, and compare it to a normal distribution).
If I could understand the logical methodology of how to distinguish random from regular, I could do the calculations.
I can't understand the methodology yet.
Generate 100-500-1000-10000 random series and check your TS on all of them - if on average the results are better or comparable to the results on the price series, then the TS should be thrown into the furnace.
Only all rows should be comparable in length to price series.
It is even possible to generate the series in Excel
Thanks Alexey, I have started to do forwards, but only on the sections of history that made it into the overall 10 year test, which has already been previously passed.
It is clear that such a forward passes the test with comparable results and apparently makes no sense.
Here it is a 1/2 forward per sample.
And there is no other (qualitative) history. Hence there is a difficulty - what to forward on?
Maybe it is possible to create synthetic quotes with properties similar to those of the tested symbol?
Then we could use them.
Then a question. Can we create such quotes that somehow inherit the properties of quotes of the tested symbol?
I can easily sample the significance of positivity. (I will add all the variance and get the average, and compare it to a normal distribution).
If I understood the logical methodology of how to distinguish random from regular, I could do the calculations.
I can't understand the methodology yet.
The point of forward testing is to trade on a part of history that did not take part in optimization. For example, you optimize a period up to the beginning of January 2018, and then look at the trade in January 2018. (using optimized parameters) and so on for each of the months. The resulting sample of 12 profits will allow you to understand how your strategy works in optimise-follow trade mode.
What you are talking about (Monte Carlo simulation) in your desired form does not seem to me applicable - we do not know and never will know the "properties" of quotes in the future. We can only do this trading simulation on random walk realizations and compare the resulting sample with the sample obtained by forward testing (goodness-of-fit criteria)
It's all empty. Even if regularities exist, it is almost impossible to find them by any algorithm.
Yes, you're right! The easiest way is to deny the possibility of a certain event ... It's so natural: you don't have to do anything, just smartly say "it doesn't exist, because - it can never be!.."
The point of forward testing is to trade on a section of history that was not involved in optimisation. For example, you optimize a period until the beginning of January 2018, and then look at the trade in January 2018. (using optimized parameters) and so on for each of the months. The resulting sample of 12 profits will allow you to understand how your strategy works in the optimise-follow trade mode.
What you are talking about (Monte Carlo simulation) in your desired form does not seem to me applicable - we do not know and never will know the "properties" of quotes in the future. We can only do this trading simulation on random walk realizations and compare the resulting sample with the sample obtained by forward testing (goodness-of-fit criteria)
Uh-huh, with the forward testing I understand... I just have to wait until that future period arrives...
Life is so short and the worm is so long)))
When I was talking about properties of quotes, I didn't mean to try to predict the future.
to inherit some unique features of the instrument, such as spread, internal oscillation range or something else...
Actually, I don't understand these features and peculiarities very well, but I see by testing that EURUSD is qualitatively different from USDCHF.
At the same algorithm settings, I get distinctly different dispersion patterns at different symbols.
Frank
Yen .
What makes them different... from each other, which means they have some characteristic properties/peculiarities.
Curious to understand - which ones, how to identify them and how to apply them in modelling to synthetics without conflicting with what you said(Random Wandering).
If you don't take these features into account, then there is no point in testing on synthetic quotes, because with the same success,
the algorithm may be tested simply on another pair...
Has the topic of specific differences of quotes by symbols been discussed/studied anywhere?
It would be interesting to read...