Machine learning in trading: theory, models, practice and algo-trading - page 3132
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
Just an observation: if you input something that describes a price chart without being bound to a strict time chronology, the results seem to be more lively or something. That is, if you input prices N+1, N+2, N+3.... then we get some kind of nonsense, a perfect 50/50 chaos in all its manifestation.
And if we feed something like N+1 , N+6, N+17... into the input. constructed according to some specific rules (for example, a graphic pattern), the results seem to give signs of life, as if the patient is alive but in a coma.
The next point: the target. If it is described as "next candle is such and such, eat neural network and update weights", then again - the output is full of randomness. If you describe the future independent of the same chronology and put marks according to some rule like "I like the chart next, there is +1 unambiguously, eat the neural network and update your weights".
If you combine both methods, variable timeframe as input and variable timeframe as target, the results sometimes start for a while (the patient is in a coma, but is twitching his arm). For example, if you train with 2021 data, almost all of 2022 can be in +, and with a strict input/target timeline, the chatter is random. But, the same 2020 doesn't lend itself, after it 2021 is chaotic. It turns out that variable data capture some working patterns and can trade with them, and strict chronology breaks the forward at once.
I read the article, ancient as a mammoth, it is still alive on the domain people.ru, where a guy writes that if you take the increment of strict chronology and somehow compare it with the past, the correlation is 0. But if you take the data describing the price chart without strict binding to the order, the correlation with the past reaches 0.8. He concludes that pricing within a candle is random and cannot be predicted. Pricing for an indefinite period of time is predictable.
I use Edge, no vpn and it also googles all languages.
I recommend to learn about genetic programming in order to research not by poking and prodding.
In order to search for regularities not tied to time/chronology, I recommend to get acquainted with algorithms asocial rules.
As for the direction of search, everything is correct...
the market has only prices and a sequence of certain events, exactly events.
I'm researching what to feed to the input of the neural network. Since you don't have a ready-made product in the market, I'll say something:
I've got a mac, I'm too lazy to bother with Edge.
Is it just for windows or something?
and it's just for Windows or something?
and it's only for Windows or what?
In one of the articles I analysed the results depending on the fixed forecast horizon. The best one was around 15 bars ahead on the hourly indicators, it guessed the direction most accurately. But there can be different periods for each instrument.
I recommend to learn about genetic programming in order not to investigate by poking and prodding.
To look for patterns that are not tied to time/chronology, I recommend learning about algorithms such as associative rules.
And the direction of the search is correct.
there are only prices and a sequence of certain events on the market.
Thank you.