Machine learning in trading: theory, models, practice and algo-trading - page 3262
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 level becomes a level when it has ended and there will be a different level in the future in which trading decisions are made, i.e. a level has no predictive power.
But in reality it is even worse.
Building TS on levels at the stage of training, testing can give excellent results because of "looking ahead", as we teach on ready-made levels, and when we start counting the predictor, there will be no level - see above. Therefore, the prediction error on different there ALE OOS can give excellent results, but if we start chasing step by step outside the training files, doing predictor calculations, the classification error will be arbitrary.
disagree with any of the above
No one has ever built a successful algorithm on stochastics and other curveballs.
This is very controversial. At least my robots refute it. But there is specificity there and some addition to "stochastics and other curveballs".
but together these strategies generate stable profits
I agree with that, that's how they work for me.
But it would be better to put it all in a separate topic, it doesn't belong here. And me too, I am not mature enough for MO yet, and there is no incentive - it works quite well as it is.
I agree with that, that's how they work for me.
Strategies can be the most unsightly, the only thing that matters is that the average balance increment is greater than zero
generate 100 conditional strategies with an average above zero
profit on individual strategies is below zero and the strategies themselves look terrible
but together these strategies generate stable profits
========================================
That's how it is... by rethinking everything a bit and giving up quality (one good TS) you can move to quantity and as a result come to quality.
=========================================
The only question is how to select strategies with non-zero expectations.
Everything else is solvable
Why then is the trading systems league not working very well ?
kind of hints :-)
Why then the league of trading systems does not work very well?
The main condition is not fulfilled there - all systems must be profitable.
On what?)
that a generally good idea (group growth/decline of typical algorithms like market indication and "drift" of their optimal parameters) was killed by an optimiser.
and the sum of strategies or a selection of strategies does not give a plus. "there are barrels of tartar and honey: no matter how you put them together and select them, they will not be better; even mixing the individual ones, you cannot separate the honey".
The main condition is not fulfilled there - all systems must be profitable.
I think that's exactly the point.
Yes, there may be one or two lousy sheep in the flock, with a slight disadvantage. But not for long, selection does its job. I'm talking about my own.