Machine learning in trading: theory, models, practice and algo-trading - page 3218

 
Maxim Dmitrievsky #:

Advanced resampling in the face of GMM other generative models do the job well.

I obtained synthetic feature values from the original ones, trained the model on them, and it worked on the original data.

There's a whole video on Renaissance on how they generated data if there wasn't enough data.

 
Valeriy Yastremskiy #:

There's a whole video on Renaissance on how they generated the data if there wasn't enough of it.

Where
 
Maxim Dmitrievsky #:
Where

https://www.youtube.com/watch?v=K10PVDm0LVw

This is the last one, but the one above is nice too, some thoughts))))))

Here))

Exposing Jim Simons Cryptic Data Tactics and Simulations
Exposing Jim Simons Cryptic Data Tactics and Simulations
  • 2023.06.16
  • www.youtube.com
Inspired form the book about Jim Simons “The man who solved the market” and how they simulated or created data to perform quantitative analysis we discuss in...
 
Valeriy Yastremskiy #:

https://www.youtube.com/watch?v=K10PVDm0LVw

This is the last one, but the one above is nice too, some thoughts))))))

Here))

Well, there's Monte Carlo.

 
Maxim Dmitrievsky #:

you know, Monte Carlo.

it looks like it's late '90s.)

take a row and noise it, first thing that comes to mind).

Get a mathematical model by removing the noise, and noise it differently.

What other ideas can there be, if the task is to get approximately the same series, but different).

 
In trade simulation, I see two ways.

1. You can modify the TS in relation to the data
2. You can modify the data relative to the TS.

In principle, nothing prevents you from combining the two.

De Prado chose the first method in his article on retraining, Saber chose the second.

I would suggest to first compare these two methods, choose the best one, and then dig into the details of a particular implementation.

The first way seems more correct to me, because we understand the parameters of the TS and they are objective, but there are a lot of uncertainties in the issue of price simulation
 
mytarmailS #:
In trade simulation, I see two ways to do it.

1. You can modify the TS in relation to the data
2. You can modify the data with respect to the TS.

In principle, there is nothing to prevent merging.

De Prado chose the first way in his article on retraining, Saber chose the second.

I would suggest to compare these two methods, choose the best one, and then dig into the details of a particular implementation.

The first way seems more correct to me, as we understand the TC parameters and they are objective, but there are a lot of questions about price simulation

It is a task to change the data when there is not enough data. And for a more complete understanding of the TS operation. Besides, stress tests require certain data, which may not be at hand.

 
Valeriy Yastremskiy #:

It is a task to change the data when there is not enough data. And for a more complete understanding of the TC operation. Besides, stress tests require certain data, which may not be at hand.

We don't have a problem with a lack of data
 
mytarmailS #:
We don't have a data scarcity problem

There is an arima.sim that models arima parameters .

And for other functions, I can't think of any. Do you know of any others? For MO functions? If they are not in the packages in R, you don't have to do this, but if they are, you can do it ready-made.

 
СанСаныч Фоменко #:

There is an arima .sim in which the arima parameters are modelled.

I can't think of any for other functions. Do you know of any others? For MO functions? If they are not in the packages in R, you don't have to do this, but if they are, you can do it ready-made.

There are a lot of things, Google can help.
For my TC simulation is not relevant, so I do not even want to get into this topic