[Feature Request]: Using different covariates

[zhangc927]

Describe the feature you want to propose or implement

Hello, I’ve encountered a problem. The variable y is a firm-year indicator, and x is a region-year indicator. I want to control for region fixed effects and time fixed effects. However, if I use a nonlinear machine learning method such as random forests, the interaction between region and time may remove most of the information in x when predicting it, which is clearly unreasonable. Therefore, I’d like to ask whether it’s acceptable to use different covariates for the outcome equation and the treatment equation. Alternatively, do you have a more reasonable approach? I look forward to your response. Thank you!

Propose a possible solution or implementation

No response

Did you consider alternatives to the proposed solution. If yes, please describe

No response

Comments, context or references

No response

[SvenKlaassen]

Hi @zhangc927,

generally using different covariates is not implemented directly.
A possible option would be to use external predictions, which are available for most models. This would allow you to fit your own models previously and use different sets of covariates (just remember to use cross-fitting).

I think have an option to use different controls would be great for the package.