ForumSections

Machine learning in trading: theory, models, practice and algo-trading - page 1157

New comment
Alexander_K2:

Knowing in advance that no one will do any of what I suggest, but still.

1. Everyone knows that real BP and its increments are 98% like a stationary Laplace motion random process, except for the 2% abnormal outliers in the "heavy" tails.

Those 2% are the reason for the 98% loss) Think of it as an enhanced version of the 80/20 rule

Aleksey Nikolayev:

That 2% is the reason for the 98% loss) Think of it as an enhanced version of the 80/20 rule

I agree.

So that's the problem - by all means remove, exclude these 2% from BP. For pure stationary Laplace motion, using Lyapunov's TSP, my TS gives the most confident "+".

Alexander_K2:

I agree.

So that is the task - by all means remove, exclude these 2% from BP. For pure stationary Laplace motion, using Lyapunov's TSP, my TS gives the most confident "+".

You'd better think about what to do with this 2%. You can exclude them from calculations, but in reality they were and will remain. "They won't hit on the passport, but on the face.
Alexander_K2:

I agree.

So that is the task - by all means remove, exclude these 2% from BP. For on pure stationary Laplace motion, using Lyapunov's TSP, my TS gives the most confident "+".

2%? Getting in the way? This is from the cycle of the bad dancer.

A 2% noise level for most signals is like no noise at all.

Yuriy Asaulenko:

2%? Getting in the way? This is from the cycle of the bad dancer.

A 2% noise level for most signals is like no noise at all.

Yuri, please show us the original BP and the converted stationary BP. You can leave the filter as a secret. А?

Alexander_K2:

Yuri, please show us the original BP and the converted stationary BP. You can leave the filter a secret. А?

I do not do this. And it's not in my plans).

Alexander_K2:

This is an artificial series of Laplace motion with the start at 0 and nothing more.

So far my TS is giving the conditional +175.86 pips on the 1st series of 25 trades (+14/-11). I'll check it further...

Can the NS get anything out of them?

I'm sorry, but I can't work purely with numbers without logic, and I'm not interested in looking for logic in excel. I'll watch from the outside.

Aleksey Vyazmikin:

I'm sorry, but I can't work purely with numbers without logic, and I'm not interested in looking for logic in Exel. I will observe from the sidelines.

There's nothing to observe. On those stationary rows, which I have posted, my TS gives the most confident profit.

Therefore, for me, the main task is to convert the real BP to a stationary form.

I think the same should be in the first place for neural networkers. Everything else is a trifle.

Alexander_K2:

There's nothing to see here. On those stationary series, which I posted, my TS gives the most confident profit.

Therefore, for me, the main task is to convert the real BP to a stationary form.

I think the same should be in the first place for neural networkers. Everything else is nonsense.

I don't know, I'm not a deep expert in neural networks, rather work with trees, but my file for finding solutions looks like a set of predictors with a target.

Is anyone working with anything other than time series/increments?

I need help, I have predictors, I have targets, but I don't know how to get the most out of the data - very little training experience (different models show different results with default settings).

Can anyone help on this issue?

1...115011511152115311541155115611571158115911601161116211631164...3749
New comment