Machine learning in trading: theory, models, practice and algo-trading - page 3497

 
mytarmailS #:

I conducted an experiment simulating a trained trading model (below with arrows/transactions)

(on the left is the original model on the right is the effect of changes on the model)



Effect of changing the linear trend


in essence, the linear trend does not affect the trained TS (but is there a linear trend in the market)



Influence of fluctuation amplitudes

Changes in amplitudes affect the TC



Effect of phase shift


Phase shifting affects the TC


Effect of frequency offset

Frequency offset affects TC

Yeah, that's cool. I've thrown some links to different augmentation methods in my channel, last posts.
Later I'll show examples of what I got.
 
Maxim Dmitrievsky #:
The second option, yes.
Lots of ways on how to do it. I even tried an extreme one, like the V-M function :)

I'll look at the achievements with interest. What is this function, Weierstrass-Mandelbrot?

Maxim Dmitrievsky #:
for each BP you need to pick something different.

This is where the magic begins, which is difficult to re-create.

Maxim Dmitrievsky #:
If stability is found and verified through cv etc., little doubt remains.

Did you do your cross-validation to take only the original data, or are you checking on the transformed data?

 
mytarmailS #:
or at least linear trend modelling.

Is there any way to significantly reduce the number of gaps?

 
mytarmailS #:
Changes in amplitude affect the TS.

This is just the volatility - so we need predictors to offset the influence.

Something to think about.

mytarmailS #:
the whole code of the experiment ))))

Concise. Is there model training and predictors in there too? Or just graph generation?

 
Aleksey Vyazmikin #:

I'll look with interest at the achievements. What is this function, Weierstrass-Mandelbrot?

This is where the magic starts, which can be difficult to re-create.

Did you do your cross-validation to take only the original data, or are you testing on the transformed data?

Yeah, in-mandelbrot.
You have to kind of look at the raw data and by mental extrapolation see what might be missing in the future, add that in. It also depends on how the trader trades.
Test on new initial data only. On supplemented + initial training.
 
Aleksey Vyazmikin #:

Is there any way to significantly reduce the number of gaps?

What gaps?
 
Aleksey Vyazmikin #:

This is just the volatility - so we need predictors to offset the impact.

Something to think about.

That's concise. Is there model training and predictors? Or just graph generation?

Yeah, it's a lot to think about.

The model is imaginary, its trading as if on the traine on the left, and on the right is a test of new data.

It's not a problem to train a real model.
 
Maxim Dmitrievsky #:
Test on new baseline data only. On augmented + baseline training.

And I wouldn't do cv on the mixture - who knows, maybe there the error is reduced only on artificial data... in any case I would test this hypothesis.

 
mytarmailS #:
What gaps?

Well, like what, between the closing and opening prices. It's like it's illiquid.

 
Aleksey Vyazmikin #:

And I wouldn't do cv on the mixture - who knows, maybe there error is reduced there only on artificial data... I'd test that hypothesis anyway.

If they're peeking, yes. They shouldn't.