Machine learning in trading: theory, models, practice and algo-trading - page 2634
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
A little more quasi-philosophy. The trade-off between bias and variance will always limit the complexity of the model. Therefore, one can never be sure that the model will work well on the entire set of predictors. Accordingly, the task arises of determining a working subset of this model. If I understand correctly, this is exactly what Maxim wrote about recently (about two models). It's quite in line with the old idea that "you shouldn't try to be in the market all the time".
It would be nice to try to combine all this into one model. For example, this idea (Aleksey Vyazmikin had a slightly similar one) -- break each predictor into segments, which gives a breakdown of the whole set of predictors into multidimensional cubes. Then from all these cubes we choose a set of suitable cubes. With large dimensionality this problem will be combinatorially intractable, but we can do by analogy with random forrest -- randomly choose low-dimensional sets of predictors. The initial segmentation for each predictor can be done by dividing the equity (when transactions are sorted not by time, but by a given predictor) into monotonic chunks.
Supplement it all with crossvalidation (forward) and other stuff) It may even be not quite nonsense) Well, or something similar has been done before.
https://habr.com/ru/company/ods/blog/544208/
Useful article.
https://habr.com/ru/company/ods/blog/544208/
Hat. A hodgepodge of different statistical tests
As a topic for reflection/understanding, that
Correlation != Causation
1-4.
And do your own tests anyway. And so the article is practically advertising)
As a topic for reflection/understanding, that
Correlation != Causation
1-4.
And do your own tests anyway. And so the article is practically advertising)
I wonder if it is possible to look for patterns in synthetic data. Let me explain - from a small sample of 100-200 observations to draw from their distribution a lot of synthetic data and there already look for some complex sequences, etc.
Logically, no.
Why?
logically :-)
PS/ too little data and they are neutered to ohlc or separate metrics altogether
PPS/ but if you take regular such samples, it's a different story. Look/search - here's a bunch of patterns, what do they have in common. Here the size of each fragment may not be large and their total number. Because the main pattern you have indicated by forming a set of