Discussion of article "The market and the physics of its global patterns" - 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
You can check with the Weierstrass-Mandelbrot ph-i, e.g., as here
The goal is to check whether the algo can find regularities where they clearly exist. If yes, then rotate quotes in different ways, such as filters by time, seasonal patterns, and so on. I.e. limit the search space.
I recently tested evolutionary programming functions on Weierstrass - gene expression clicks like nuts, picks up formulas in 20-40 seconds, I haven't figured out symbolic regression yet.
there is a small problem - one out of 5-7 runs of genetics can "go wrong" and find nothing at all, which is logical in principle - run with random values - mutations and crossover can pick up unsuccessful combinations of genes, but it works very fast, with BP, in general, it works adequately, but if you run it on forwards, the number of successful runs drops sharply, but they seem to be there.
ZY: about robots - only recently on the radio "Humour FM" heard an old number of A.Revva, words there are not many; "these cyborgs ... they have filled, they have filled everything...". ))) - on YouTube should be
I recently checked the Weierstrass functions of evolutionary programming - gene expression, like nuts, it picks up formulas in 20-40 seconds, I haven't figured out symbolic regression yet.
there is a small problem - one out of 5-7 runs of genetics can "go wrong" and find nothing at all, which is logical in principle - start with random values - mutations and crossover can pick up unsuccessful combinations of genes, but it works very fast, with BP, in general, it works adequately, but if you run it on forwards, the number of successful runs drops sharply, but they seem to be there.
ZY: about robots - only recently on the radio "Humour FM" heard an old number of A.Revva, words there are not many; "these cyborgs ... they have filled, they have filled everything...". ))) - on YouTube should be
Genetics is more fitting, there are no regularisations and so on. Cutbust, for example, won't train well (in terms of not memorising) a sequence if the regularity is weak. I.e. a tree or forest will be perfectly memorised, but with bousting this will not work. More confidence in the models.
Regarding symbolic regression and so on: as far as I know, it is an outdated algorithm, there are better ones. Here are the latest innovations for time-series classification. I haven't used it, some of them take a long time to learn (a lot of overshooting). The whole list is in the "algorithms" tab. Tests, and other interesting things.
Yeah, that's a good bruteforce. I do this myself, only through machine learning models. Often, patterns are found better if you limit the search to specific hours, i.e. open trades only at certain times. For example, there is a different optimisation for each hour of the day.
In the market there is a popular bot from one author (in the top), he picked up such filters that only the beginning of the London session is traded. I.e. trades only at the opening of the session. Passes tests for 20 years. It's cool. I don't have such a long time, but it's not bad either.
Yes, I've seen such bots, that's where I got the idea from. I didn't even realise that time corridors could do that. The corridor does not even have to coincide with sessions, as Bruteforce shows, there can be all sorts of completely unexpected time segments. One such segment is about half an hour to an hour to the left and right of 0:00 (day change point). It's a very cool segment. Only there with spreads you need to be very careful ) .By the way I wanted to say that tests show that let's say if you take a sample of say 10 years, then on the forward it all works for another year or so. I did not look further into the future. Such Expert Advisors are extremely simple, but take for example a weekend when the stock exchange stands, generate a dozen of them for each currency pair and take profit for half a year. Even despite the simplicity of such robots, they can be predicted, but an owl written by hand will not be able to boast of this. I am already beginning to think that it is better to entrust a machine to write owls )) to trade for half a year and then to trade some more )). As it seems to me it is only a matter of time before the machine starts to do it all better and faster than we do, yes it already is, to be honest.
Yeah, I've seen bots like that, that's where I got the idea from. I didn't even think before that time corridors could do that. The corridor does not even have to coincide with sessions, as Bruteforce shows, there can be all sorts of completely unexpected time segments. One such segment is about half an hour to an hour to the left and right of 0:00 (day change point). It's a very cool segment. Only there with spreads you need to be very careful ) .By the way I wanted to say that tests show that let's say if you take a sample of say 10 years, then on the forward it all works for another year or so. I did not look further into the future. Such Expert Advisors are extremely simple, but take for example a weekend when the stock exchange stands, generate a dozen of them for each currency pair and take profit for half a year. Even despite the simplicity of such robots, they can be predicted, but an owl written by hand will not be able to boast of this. I am already beginning to think that it is better to entrust a machine to write owls )) to trade for half a year and then to trade some more )). As it seems to me it is only a matter of time when the machine will do it all better and faster than us, yes, it is already there to be honest
... in general, the tester does not correspond to reality.
Of course, but if there is a result in the tester, there are more chances to get a working strategy in real life. There is one more caveat, if the robot is designed for high timeframes and shows a mathematical expectation much higher than the average spread, such tests will be absolutely objective. You can see an example in this article for example https://www.mql5.com/en/articles/8767.
... the tester doesn't correspond to reality.
What's all this got to do with physics, mate?
What's all this got to do with physics, mate?
Really nothing, so mathematicians and physicists come here, just because they are so fed up with these laws of conservation of momentum and energy that they are even ready to write such things. Where is science going, where are our bases on Mars? ) Where are our Alcubierre engines ? )