Machine learning in trading: theory, models, practice and algo-trading - page 1436

 
Alexander_K:

:)))) Kesha is SanSanych's grandson, on whose pension he survives. Alesha is the ears of investors, Uchitel is at the car wash, and Koldun is in the woods in the bathhouse. Who else? I can't say anything about Doc. I don't know.

No sign of him on the English-language thread, either? I'll go and check.

Aleshu investors executed, he is dead, as well as Yura Reshetov, enough here sacrilege and dance on the graves, forget about these names, try to find the grail yourself, now no one will help.

 
Alexander_K:

I am also genuinely perplexed.

I went to a forum of experts like Prival and Matemat, and came to a forum of Alesh and Indians. It's a paradox!

Prival has long outgrown Forex spiritually, grows tomatoes at his dacha outside Moscow and remembers trading only in his nightmares. Mathematician is drunk.

 
govich:

Why are you torturing the software from the last century, on the cyberforum offered a variant five times faster.

Cool, but it's still a bit complicated for me. I guess I've tried to guess a little bit of Neuropro and came to a conclusion that it cannot find regularities in market quotes even if they really exist, because it is adjusted in training area for maximal profit (that is the percentage of correct forecasts).

As a result the grid will hammer trades on every bar, while in reality it should skip some trades when we need to trade some pattern.

 
Примеры проектов R - Visual Studio
Примеры проектов R - Visual Studio
  • 2018.01.24
  • kraigb
  • docs.microsoft.com
Индекс коллекции примеров для начала работы с R и Visual Studio.
 

Also a variant. Actually, the toolkit is not so important, it is possible to implement it in R. Another thing is that as far as I understood machine learning works fine on real data, but not so much on quotes.

 
I think it would not be a bad idea to put the time between two consecutive events into the neural network along with the price (return) of the thinned tick series.
 
Alexander_K:
I think it would not be unreasonable along with the price (return) of the thinned tick series to put into the neural network the time between two consecutive events.

I think you are too much into thinning ticks. In fact, you are not catching any additional HFT information, but the specifics of aggregation-filtering-distortion from individual DCs. Besides, pure ticks(transaction price) are not very good as a price indicator, because they have a strong backward autocorrelation of the first lag, at high frequencies, for the obvious reason (the price beats then in the bid and ask). For more uniform price in HFT, take the average bid\ask from the cup, or even better a weighted average of the volume (when it is) in the cup.

 
Alexander_K:
I think it would be nice if along with the price (return) of a thinned tick series, the time between two consecutive events would be added to the neural network.

no returnees from now on, in my personal message I've sent you the best way, read it at your leisure )

 
Maxim Dmitrievsky:

No returnees from now on, I've emailed you the best way, read it at your leisure.)

Yeah. I read it little by little.

 
Alexander_K:

Yeah. I'm reading a little bit.

At least in that process I saw both stationarity and the presence of mutual information with the original row. There are some outliers, which can also somehow be fixed, but you'll see.

the formula is simple, I rewrote it on mql