How many features is too many for a financial ML model? #20
Answered
by
martex-dev
martex-dev
asked this question in
Q&A
|
How many features is too many for a financial ML model? Notes and answer below. |
Answered by
martex-dev
Aug 25, 2026
Replies: 1 comment
|
The signal-to-noise ratio in financial returns is low enough that the practical limit is far lower than in most domains. A useful discipline is to require every feature to have an economic reason for existing before it is allowed into the model, then check whether it survives a purged K-fold with an embargo. Features that only look good under standard K-fold are usually exploiting serial correlation across the fold boundary. |
0 replies
Answer selected by
martex-dev
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The signal-to-noise ratio in financial returns is low enough that the practical limit is far lower than in most domains. A useful discipline is to require every feature to have an economic reason for existing before it is allowed into the model, then check whether it survives a purged K-fold with an embargo. Features that only look good under standard K-fold are usually exploiting serial correlation across the fold boundary.