Machine learning in trading: theory, models, practice and algo-trading - page 2794

 
СанСаныч Фоменко #:

So they are correlated! The result depends on the order in which the features are discarded.

You've sifted down to 75%.... Back to your packages again... how are they better than the real model?
 

With such frequent retraining at each bar, randomisation in model training (random splits) will have a strong influence. It can bury the rest of the alpha and it will

because Murphy's Law says Anything that can go wrong will go wrong.

Inefficiencies have to be literally ripped out with a scalpel, and models tend to flatten and average out, leaving nothing to pick up on.
 
Maxim Dmitrievsky #:

You can fit 500 bars, no doubt, if you predict for 1 trade. You need statistics, I do not believe that it will predict better than random on average. But the variant has the right to life.

regarding the selection of features and multicollinearity (I specifically asked a question to the developers of bousting) - it makes sense to select only in case of contests, to get cleaned models and in the struggle for fractions of %. In all other cases it makes almost no sense to do such preprocessing. They zero out the rubbish perfectly well by themselves.

1. Clarification. The 500 bar is a statistic for selecting chips by predictive power. It is not a model statistic. I did the model statistics on 1000 bars, retraining the model on each bar. Here the prediction error is always above 80%.

2. Multicollinearity. As a fact on my chips. I see no reason to generalise. But a curious result. The given value of 75% is some optimum, the middle between 70% and 80%. Above and below these values the prediction error is larger relative to 75% by about 10% difference. With a total error of no more than 20%, that's a lot.

 
elibrarius #:
You've sifted it down to 75%..... Back to your packages again...how are they better than the actual model?

Let's not confuse the foundation with the whole building. But without a foundation, a building can collapse.

 
СанСаныч Фоменко #:

1. Clarification. The 500 bar is a statistic for chip selection on predictive ability. It is not a model statistic. I did the model statistics on 1000 bars, retraining the model on each bar. Here the prediction error is always above 80%.

2. Multicollinearity. As a fact on my chips. I see no reason to generalise. But a curious result. The given value of 75% is some optimum, the middle between 70% and 80%. Above and below these values the prediction error is larger relative to 75% by about 10% difference. With a total error of no more than 20%, that's a lot.

SanSanych, please have a moment of patience. There is no one else to ask. All the MO experts are all honourable guys and won't even talk to me).

I am curious, what do you want to predict with MO?

1. The next tick?

2. the next bar?

3. The next group of ticks?

4. Thenext group of bars?

5. The direction of the wave?

6. The direction of thetrend?

It seems to me that the Tourettes crowd has no idea what they want to find.

It seems that they want to find money signs at once).

Can you at least explain it in a nutshell? For the exact purpose of the MO, so to speak.

 
Oh, I'm all excited to hear back
 
Uladzimir Izerski #:

San Sanych, please be patient for a moment. There is no one else to ask. All the MO specialists are horny guys, and they won't even talk to me.)

I am curious, what do you want to predict with the help of MO?

1. The next tick?

2. The next bar?

3. next group of ticks?

4. Next group ofbars?

5. The direction of the wave?

6. Direction of thetrend?

It seems to me that the Tourettes crowd doesn't even have an idea of what they want to find.

The feeling is that they want to find money signs at once).

Can you at least explain it in a nutshell? For the exact purpose of the MO, so to speak.

I predict the next bar on H1. Prediction result = -1; 0; 1.


But that's on the surface. There is a whole internal complexity. I will not specify. Think for yourself.

To formulate a teacher correctly is no less important than to match predictors that are related to such a teacher.

 
mytarmailS #:
Oh, I'm all excited for the response

I was expecting that response. You won't believe it.))))

=======

I would like to hear from you and other MO dreamers.

But I guess I will never get it. After all, you yourself do not know what you are looking for))))))

If you tell me, maybe I will help in some questions.

P.s.

If no one knows the answer to my question, then no one knows what they are looking for.

Here is the ANSWER. You can refute it.

 
Uladzimir Izerski #:

I was expecting that reaction. You won't believe it.)))

I do

Uladzimir Izerski #:

I would like to hear a response from you as well

Personally, I like to predict bounces/extrema, and many other things....

In fact, the answer lies on the surface - "what you programme you predict".

Uladzimir Izerski #:

I would like to hear an answer from you and from other MO dreamers.

But apparently I will never wait for it. After all, you do not know what you are looking for))))

question : what are you looking for??? and immediately the statement : you yourself do not know what you are looking for!!!

:)))))) funny, sharp, thoughtful )

Uladzimir Izerski #:

If no one knows the answer to my question, then no one knows what they are looking for.

if you have a million, but you don't want to give me half of it, then you don't have a million.

:))))) funny, sharp, thoughtful )

 
mytarmailS #:

I believe

Personally, I like to predict bounces/extremes, and many other things....

Actually, the answer lies on the surface - "what you programme, you predict".

the question: what are you looking for??? and then the statement: you don't know what you're looking for!!!

:))))) funny, sharp, thoughtful )

If you have a million and you don't want to give me half of it, then you don't have a million.

:))))) funny, sharp, thoughtful )

What can I say?

Sandbox!

No specifics).

Goodbye.