Machine learning in trading: theory, models, practice and algo-trading - page 3691

New comment
 

Almost a complete reversal of one top market strategy on gold using MO. :)


 

Sometimes the markup problem for trending TCs is solved through this approach.

https:// www.mql5.com/ru/articles/17654

Машинное обучение в однонаправленной трендовой торговле на примере золота
Машинное обучение в однонаправленной трендовой торговле на примере золота
  • www.mql5.com
В данной статье рассматривается подход к торговле только в выбранном направлении (на покупку или на продажу). Для этого используется техника причинно-следственного вывода и машинное обучение.
 
Enjoyed the article. It will be possible to try to attach machine learning to the strategy.
Statistical Arbitrage Through Mean Reversion in Pairs Trading: Beating the Market by Math
Statistical Arbitrage Through Mean Reversion in Pairs Trading: Beating the Market by Math
  • www.mql5.com
This article describes the fundamentals of portfolio-level statistical arbitrage. Its goal is to facilitate the understanding of the principles of statistical arbitrage to readers without deep math knowledge and propose a starting point conceptual framework. The article includes a working Expert Advisor, some notes about its one-year backtest, and the respective backtest configuration settings (.ini file) for the reproduction of the experiment.
 
Maxim Dmitrievsky #:
Enjoyed the article. We could try to put machine learning into the strategy.

in the title arbitrage, does it make sense? I remember your posts from a long time ago, you still can't part with it?

 
Maxim Dmitrievsky #:
Enjoyed the article. Might try to screw machine learning into the strategy.
It would be interesting to add macroeconomics to the signs.
 
Aleksey Nikolayev #:
It would be interesting to add macroeconomics to the recognitions.
Interesting article. Yes, more examples like this. You can't immediately figure out what to screw where.
 

It says '' The spread between correlated pairs tends to revert to the mean.'' ordinary differential equation (ode) solvers can work well theoretically if the underlying dynamical equation if time series is known.
https://github.com/rtqichen/torchdiffeq ; it is complicated but gemini and chatgpt give some basic examples .

GitHub - rtqichen/torchdiffeq: Differentiable ODE solvers with full GPU support and O(1)-memory backpropagation.
GitHub - rtqichen/torchdiffeq: Differentiable ODE solvers with full GPU support and O(1)-memory backpropagation.
  • rtqichen
  • github.com
This library provides ordinary differential equation (ODE) solvers implemented in PyTorch. Backpropagation through ODE solutions is supported using the adjoint method for constant memory cost. For usage of ODE solvers in deep learning applications, see reference [1]. As the solvers are implemented in PyTorch, algorithms in this repository are...
 

To return to the mean, we need cointegration, not correlation. Correlation between exchange rate increments is always present, and it greatly hinders to determine the real presence of cointegration in tests. When correlation is searched for between prices and not their increments, it is essentially a search for cointegration - cointegration is searched for through the construction of a regression of prices, which is determined through correlation.

In financial mathematics, not ODEs but SDs - stochastic diffusions - are usually used for modelling series. But the point is the same - if the exact equation of the exchange rate series were known, everyone would have become a trillionaire long ago. More precisely, if the price was really given by an equation (no matter ODE or SDU), it would be quickly determined by mathematicians and would become a source of grail. When they try to pull it off in practice, everything is as usual - the equation works fine on the fitting history, but on the OOS - sadness, ennui and factory pipes on the horizon.

 
In pieces.app added claude 3.7 for coding, you can use it for free without vpn. According to first impressions, it offers more interesting options than the old 3.5. For example, it is better with kozool inference. For python.

The old version, from my code snippet, didn't realise it was kozool. This one did.
 
Aleksey Nikolayev grail. When they try to pull it off in practice, everything is as usual - the equation works fine on the fitting history, but on the OOS - sadness, ennui and factory pipes on the horizon.

Well, we proceed from the non-mathematical term that if at least a month "zbs" works on new data, it is already bread without physical labour :)

1...368436853686368736883689369036913692369336943695369636973698...3700
New comment