Machine learning in trading: theory, models, practice and algo-trading - page 2468

[Evgeniy Ilin]

Interesting thread, though almost useless, so much information, but almost all useless. Who moves the prices, why and why all these questions are unnecessary, as well as what data to use for analysis. Neural networks, machine learning, God ... First you need to understand what you're looking for.

[mytarmailS]

Evgeniy Ilin #:
Interesting branch, though almost useless, so much information but almost all useless. Who moves the prices, why and what for all these questions are superfluous, as well as what data to use for analysis. Neural networks, machine learning, God ... First you need to understand what you're looking for.

How?

[JeeyCi]

elibrarius #:
A picture of 3-5 deals tells you nothing. 100-200 real trades made (or in 3-4 months) will show the statistics. And the drawdowns.

Why do you mix a neural network and money-management? A neural network will give you a signal (if it is correctly trained - and this is only a probability that what it was trained on, still moves the market), but how you attach your money-management & trade-management to it (an indicator with a signal) - this is your personal attitude towards risk

p.s. Evgeny Dyuka thanks for yourlink to the example, but the logic inside the code is always more interesting)

[SanAlex]

mytarmailS #:

How's that?

How can you understand you here? - Five years in this thread and nothing has been demonstrated.

[Forester]

JeeyCi #:

Why do you mix a neural network and money-management?... the neural network will give you the signal (if it has been trained correctly - and this is only a probability that what it has learned is still moving the market), but how you attach your money-management & trade-management to it (the indicator with the signal) is your personal attitude towards risk

p.s. Evgeny Dyuka, thank you for yourlink to the example, but the logic inside your code is always more interesting)

You guys go ahead and take a look. If you unconditionally believe this indicator.
Signal it.

We'll see...

[Evgeny Dyuka]

elibrarius #:
You go ahead. If you unconditionally believe this indicator.
Signal it.

And we'll see...

What's your problem?
If you go to the first top 5 indicators and see whether there is a requirement in the comments to first trade for 3 months and then post.
If you have something that does not work, do not troll.

[JeeyCi]

Maxim Dmitrievsky #:
Here we have such a scheme, or rather, there can be two of them.

...

2. Search strategy through any classifier, analyze its internal structure (feature importance, shap values and different metrics). It is possible to automate, approaching to some semblance of AI. The output is a black box, but in the hope that the selection criteria work.

Recurrence networks and Bayesian methods, by themselves, have not demonstrated either the ability to pull "memory" out of financial time series or to get conclusions about the most robust model on new data.

Yes there is something - I'll look through it, thanks... or here

[Forester]

Evgeny Dyuka #:
Do you have some problems?
If you go to the first top 5 indicators and see if there is a requirement in the comments to first trade for 3 months, and then to put them out.
The market is more suitable for such indicators than the market itself.

Do you succeed with your indicator?

[JeeyCi]

elibrarius #:
Does it work for you with your indicator?

here you clearly missed the "Fear and Greed" theme at the beginning of those 5 years... probably because you choose recurrence development instead of incremental development...?

[mytarmailS]

No one is interested in a chart of average positions of traders at 10 brokers?

No one is interested in the fact that the price has an inverse correlation with the mood of traders?

cor(cl,av)

[1]

-0.74

Doesn't anyone interested in the fact that it explains why models don't work with new data, why traders are always losing, why the mechanics of the market reveals... no ?