Machine learning in trading: theory, models, practice and algo-trading - page 2555

 
mytarmailS #:
So why doesn't the class predict correctly? Because the quotes are not what the model expects, wrong distribution. And if you generate quotes from the correct distribution, it would be ok...

I don't remember if I did it or not, I had a similar idea

But most likely there is no connection between the past and the future. I.e. the future state for n bars is not so easy to predict, and may be more difficult than 1-2 steps ahead.
 
Maxim Dmitrievsky #:

Try it, I don't remember if I did it or not, I had a similar idea

but most likely there is no connection between the past and the future. I.e. the future state for n bars is not so easy to predict, and maybe even more difficult than for 1-2 steps ahead.
In theory : predicting the future distribution is almost the same as predicting the moving average of a series, and that's easier than predicting the price itself, there's more inertia... maybe I'll try tomorrow if I'm in the mood
 
Maxim Dmitrievsky #:

Assuming that in the future the model should also work ) errors of all kinds (including noise) will always be there, the task is to find a balance. Therefore we are talking about the same thing, in fact.

Actually, I was solving this problem in a different way, so I am writing leading questions

If you look at it from a global point of view, I agree with you. No matter what set of algorithms is taken in the end - the problems when using it will always be the same.

By the way, the similar idea (ejection of deals with indicator values far from the typical ones) was expressed by fxsaber in his blog today.

 
I read a book from the 70's that said that if there is no autocorrelation, no prediction is possible. Is there anything more modern on this subject?
 
Aleksey Nikolayev #:

If you look at it from a global perspective, I agree with you. No matter what set of algorithms is used in the end - the problems when using it will always be the same.

By the way, a somewhat similar idea (the release of trades with indicator values that are far from the typical ones) was expressed by fxsaber in his blog today.

I completely agree. Fluctuations in the corridor with dynamic width implies a choice of balance in the forecast and its probability.

 
Tell me a newbie, and the optimization of the EA in the tester/optimizer is the MO?
 
Sceptorist #:
Can you tell me if optimizing an EA in a tester/optimizer is MO?

Optimization is more like the process of training a neural network .

There used to be even solutions to automate optimization to get the most relevant sets.

 

A lecture on model selection

so there is no misunderstanding, so to speak


 
Sceptorist #:
Can you tell me if optimizing an EA in a tester/optimizer is MO?

MO is just a set of algorithms. For some of them, the tester optimization is good enough. For example, if you need to find the optimal K in KNN. But most algorithms have their own specific way of optimization. For example, gradient descent in neural networks.

In general, the answer depends on the ability to fit the specific model to the given optimization algorithms (brute-force and genetic optimization)

 
Maxim Dmitrievsky #:

A lecture on model selection

so there's no misunderstanding, so to speak.


Vorontsov is probably the best expert on MO in Russia. The course is therefore bound to be good, but since it is for IT people, it omits basic and important for us mathematics. I`ve noticed more than once, that for the application of mathematical methods in trading their basic, simplified form is not very suitable.

MO is based (see for example Tibshirani) on the assumption that there is a constant joint distribution of predictors and responses P(X,Y). From it, the conditional probability Py(Y|X) can be calculated, from which the regression Y=f(X) can be calculated. Eventually, this regression is approximated by some MO models. In the physical world, this theory more or less works. But not in trading. It turns out that P(X,Y) changes unpredictably with time (non-stationarity) and the whole theory collapses a bit.

The most popular approach is just to ignore non-stationarity and then be surprised with the results and complain about the MO).