Machine learning in trading: theory, models, practice and algo-trading - page 3160

 
СанСаныч Фоменко #:

Areview on the fingers, what a beast the cajual is.

The article links to another book on the subject - here's the most recent version of it.

I'm most interested in the question of what might constitute a "tritment" in the application of cajusal (not casual! 😁) to trading.

 
It's a paradigm shift. When you write but don't understand. It says your classifiers are associative models. No, we've been frantically searching for features for them and pretending nothing happened.

Just the word "associative" should knock the overcomers out, unless they're in a tank. This is, as they say, the collapse of the cabin :) and there already with or without tritment :).

And further, as you read, you will be waiting for more and more disappointments and defeats. Even on relatively simple datasets. That's why this book is for the brave and cowardly 😀
 
Aleksey Nikolayev #:

The article links to another book on the subject - here's the most recent version of it.

I'm most interested in the question of what might constitute a "tritent" in a causal (not casual!😁) application to trading.

You can take the tritment out of brackets (or bring in from behind brackets) and just use the remaining matstat :)
 
СанСаныч Фоменко #:

Areview on the fingers, what a beast the cajual is.

As I understood from reading all this Causal Inference is extremely relevant in the framework of evidence-based medicine, where it is not initially clear whether a new drug affects the disease or not. In medicinecausal Inference is the basis of its evidence .

In economics, Causal In ference in this "medical" sense is not so relevant, because economics is a well-determined process based on production and sales chains.

Quite a lot of uncertainty exists in the exchange rate with unknown reasons within days and weeks, but on longer periods the reasons become obvious. But we need to predict the future, but the reasons that influence the exchange rate in the future by the owners of instruments that influence the exchange rate are carefully hidden and we can only find out after the fact.


One last thing. From a statistical point of view there is nothing new in Causal Inference. Here is the opinion of CRAN

There are no basic R functions that are direct implementations of standard causal inference designs, but many methods - more or less complex - are implemented in different packages on CRAN, which we structure into main topics:

Maxim, as a person who neglects R, and therefore has extremely limited knowledge of tools, drives a cajual wave as some kind of discovery. This is nothing new, just clever Russian guys with a 90's flair have combined a bunch of well-known tools and hung a signboard "Causal Inference" with dubious applicability in economics.


CRAN Task View: Causal Inference
CRAN Task View: Causal Inference
  • cran.r-project.org
Overview
 

Sanych, when he tries to argue and prove, starts going through packets at X2 speed... It's like counting rosary beads to calm him down.

Ask him about one package from the list, he will give you links to packages referring to this package as a proof of his deep understanding of the topic.

 
Maxim Dmitrievsky #:
You can take the tritment out of brackets (or put it in from behind brackets) and just use the remaining matstat :)

What is interesting is the theoretical-quasi-philosophical aspect of what can be considered as a tritent) It is in the sense of practice everything is simple - what works is good)

If the entity is a patient who is given a medicine/placebo or a schoolboy who is given/not given a tablet, everything is clear and simple.

If in our case we take (for simplicity) the price increment per bar as the basic entity, then we, small players, have no tritment that affects it. If we take the increment of equity as an entity, then we can already take the size and direction of the position on the instrument as a tritent. But we can not stop there and take any parameters describing the TS, which actually calculates the position, as a tritment. This approach opens the gate to infinite complexity and flexibility for the concept of a tritment, which can lead to retraining, but probably also to something good).

 
Aleksey Nikolayev #:

The quasi-philosophical aspect of what can be considered a tritent is interesting. In terms of practice, everything is simple - what works is good).

If an entity is a patient who is given a medicine/placebo or a schoolboy who is given/not given a tablet, everything is clear and simple.

If in our case we take (for simplicity) the price increment per bar as the basic entity, then we, small players, have no tritment that affects it. If we take the increment of equity as an entity, then we can already take the size and direction of the position on the instrument as a tritent. But we can not stop there and take any parameters describing the TS, which actually calculates the position, as a tritment. This approach opens the gate to infinite complexity and flexibility for the concept of a tritment, which can lead to retraining, but probably also to something good).

That's roughly how I think of it. A tritment is an instrumental variable that should lead to something good. In kozula there is a separation into covariates and tritment only because we can't influence covariates, but we can influence tritment.

Plus covariates have a different meaning than features in forecasting. They are the distinguishing characteristics of each observation. So the ML model in kozul is more like a database that is queried by another model and the statistics are counted. Or 2 models are built and their intersections are looked for. More like working with databases.

Then a tritment can be represented as a query, the results of which are used to calculate statistics.
 
Aleksey Nikolayev #:

It's originally about the two-factor method of treatment verification (tritment). You, I remember, are close to this topic in the direct medical sense.

IMHO, Maxim somehow very broadly and creatively transfers the concept of tritment to our tasks.

Medicine is not a science)))) So there proper medical research practices that sustainable patterns of treatment are considered to have a cause, and non-sustainable ones do not have a cause, but have associations in observers))))) And the method of random experiment reduces error, but in no way removes it completely)))))

So close in meaning I understand))))))

 
Valeriy Yastremskiy #:

Medicine is not a science)))) That's why there proper medical research practices that steady patterns of treatment are considered to have a cause, and not steady patterns have no cause, but associations in observers.))) And the random experiment method reduces error, but in no way removes it completely)))))

So close in meaning I understand))))))

You can identify groups that the treatment has more effect on. And not treat others.
 
Maxim Dmitrievsky #:
It is possible to identify groups on which treatment has a greater effect. And not treat the others.

Well, that's what they do, diagnoses are groups, treatment is tritment))))) An essentially normal approach when there is insufficient data to understand the process. Assumption in the form of replacing logical evidence with hypothesis confirming experiments in different variations and comparisons with sufficient purity of experiment.