Machine learning in trading: theory, models, practice and algo-trading - page 3366
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
good book
Machine learning for factor investing
There are more and more good books on MO. This one is one of them.
I have cursorily identified what is close to me
1. The non-stationarity of time series cannot be ignored. Any formula, any algorithm must answer the question of applicability to non-stationary data.
2. it is obligatory to pay attention to the influence of predictors (predictive ability) on the target. maxim has led me into fornication with his understanding of cause-and-effect relations, and the book has put everything in its place: the book understands exclusively as I understood and wrote repeatedly about "predictive" ability, its stability.
For those with more than two neurons in their brains...
On one side is a forum critic who no one will hire even as a juniper in a DS office because he'll be rejected at once...
On the other side is the book he's criticising, with this list of references.
Perhaps realised what concept is meant. By way of example.
Let's assume that there is a channel TS, where trading goes from the borders inwards. And here I set the parameter to be optimised, which is the coefficient of the channel size (width).
The classical way is fine. You optimise it and see how it affects you.
According to the proposed concept, you cannot do it this way, because this parameter does not depend on the initial data (quotes history). In the proposed terminology, it is a "constant" optimised parameter.
Even if this parameter affects some polynomial, it is also a "constant" parameter, because there is no dependence on the VDC.
That's an interesting thought. Thanks.
Yes, through optimisation at least it is logical that you can find cheaper significant parameters somehow affecting the optimised ones at least linearly. The classic way is randomly or a complete search of other parameters and formulas. But this is a curse.
You can't escape from constants of course, even with complex feedback, the constant will be the beginning of calculations. (feedback in my current understanding is the effect of current data on past data, on the basis of which future data is calculated).
In any case, it is a search for new predictors and calculation formulas, possibly more significant than the optimised ones.
Yes, through optimisation at least it makes sense that you can find cheaper meaningful parameters
The point is not that, but that these parameters are dead from the moment they are found, because they "worked" in the past...
But if instead of a parameter (constatny) correct formula, it is adaptability, you can say a system without parameters. And at the same time it's better than if it had parameters.
Apparently, I'm far from high matter - I didn't understand anything again. You don't have to write for me.
It's not the point, it's that these parameters are dead from the moment they're found because they've worked in the past.
But if instead of a parameter (constatna) the correct formula, it is adaptability, you can say a system without parameters. And at the same time it's better than if it had parameters.
There is a trading strategy with one parameter, we can conditionally express it by the formula F(x), where F is a strategy, x is a static parameter.
If we use the dynamic parameter x instead of the static one, it means using the function Y instead of x, which will look like F(Y()).
So, how to find the function Y() without using optimisation, so that this function does not turn out to be as "dead" as the static x?
Apparently, I'm far from high matter - I didn't understand anything again. You don't have to write for me.
No, think only about the task, without distractions)))))) And the search for over_over_parameters is definitely not today.)))))
There is a trading strategy with one parameter, we can conditionally express it by the formula F(x), where F - strategy, x - static parameter.
If we use the dynamic parameter x instead of the static parameter, it means using the function Y instead of x, which will look like F(Y()).
So, how to find the function Y() without using optimisation, so that this function does not turn out to be as "dead" as the static x?
The problem is this)))))