Bayesian regression - Has anyone made an EA using this algorithm? - page 43

 
Дмитрий:
Stationarity is the property of a process not to change its characteristics over time.
What characteristics in particular?
 
Dmitry Fedoseev:
What characteristics specifically?
Dispersion
 
Дмитрий:
Dispersion
and that's it?
 
If you have the variance of a series tending towards infinity, what are you going to predict there?
 
Dmitry Fedoseev:
and that's it?
Broadly speaking, there is also an IO and a distribution function
 
Дмитрий:
In a broad sense also the MO and the distribution function
Then, in a broad sense, if MO, then stochastic is enough. No?
 
Dmitry Fedoseev:
Then, broadly speaking, if MOE, a stochastic would suffice. No?
Concentrate on variance - that's where the root of the problem is
 
Дмитрий:

Non-stationary data are not predicted by time series models. Neither statistical models (regression, autoregression, smoothing, etc.) nor structural models (NS, classification, Markov chains, etc.).

Only subject area models.

I cannot agree with you about classification.

The problem of non-stationarity is not seen there at all. Models on nominal (categorical) data are quite acceptable. Non-stationarity has nothing to do with nominal data at all. Moreover, converting random variables to nominal, e.g. RSI to levels, is very beneficial to the results.

There follows non-stationarity, a problem that is fundamental to any modelling - overfitting (overfitting) of the model. And to solve the problem of overfitting one has to seriously deal with predictors.

 
СанСаныч Фоменко:

I cannot agree with you about the classification.

There is no problem of non-stationarity there at all. Models on nominal (category) data are perfectly acceptable. Non-stationarity has nothing to do with nominal data at all. Moreover, converting random variables to nominal, e.g. RSI to levels, has a very favourable effect on the results.

There follows non-stationarity, a problem that is fundamental to any modelling - overfitting (overfitting) of the model. And to solve the problem of overfitting one has to seriously deal with predictors.

Classification is also based on the characteristics of the input data, and if these characteristics change over time, then future application of classification will produce incorrect predictions
 
Everything is sad...