Bayesian regression - Has anyone made an EA using this algorithm? - page 44
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Classification is also based on the characteristics of the incoming data, and if these characteristics change over time, then future application of the classification will give an incorrect prediction
It's all sad...
Exactly. Decomposing the data into bins (pockets) is easy. The problem will be when the probability distribution on the bins changes on the data outside the training sample.
Nothing lasts forever under the moon.
But there is something about classification that is very close to the trader's ear.
We sit and stare at the charts and try to find some patterns. And here it is: the intersection of two bars! Not to mention such a pattern as "head and shoulders".
And then you run the algorithm and it finds several hundreds of trees (a hundred times more happiness than with mashes), which are combinations of input data values, which can be associated with the output variable. Just a kinship of souls and TA, but at what level!
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
Obviously I don't understand something, but what statistics and what models we should talk about if the correlation function for a derivative of a Wiener process is a delta function. Of course market data is not a Wiener process in its pure form (at least not in a homogeneous stationary environment), but the correlations in today's market are significant on an interval usually not more than 1-2 hours, mostly somewhere between 15-30 minutes. And, actually, it is not a fact that this is reality and not a "seeming reflection of the seeming moon" (c)
It is interesting that your point of view strongly coincides with mine ) I have shown the presence of stable "correlations" or better said dependencies on several predictors just for the interval from 20 minutes to an hour. Read:https://www.mql5.com/ru/blogs/post/661499
But this is not yet the final truth. The significance of probability skew for a Boolean variable (predicting the sign of price movement) exists at more distant horizons. I'll write about this in more detail.
Nothing lasts forever under the moon.
But there is something about classification that is very close to the trader's ear.
We sit and stare at the charts and try to find some patterns. And here it is: the intersection of two bars! Not to mention such a pattern as "head and shoulders".
And then you run the algorithm and it finds several hundreds of trees (a hundred times more happiness than with mashes), which are combinations of input data values, which can be associated with the output variable. Just a kinship of souls and TA, but at what level!
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
Where can I read about "subject area models"? More precisely, as far as I understand, in relation to the subject area "price quotes/exchange price quotes".
Applied is fundamental analysis.
Subject matter models are models that explain a process by factors outside the time series. For example, thermodynamics