Cross-Validation function using Exogenous features

[theforecastnoob]

I have a question.
If using the StatsForecast Cross-Validation function with e.g. ARIMAX, what does the function do with the exogenous data?
To use exogenous variables, we need to provide the model with exogenous- actuals and exogenous-forecasts.

To my question, does the Cross-Validation function produce forecasts for the exogenous within the folds, or does it use the exogenous actuals as the "forecasted" period, within the folds?

[jmoralez]

Hey. It's not exogenous-forecasts, it's supposed to be the actual values of your exogenous features, which you can decide to forecast yourself. So what the CV method does is use the actual values.

[theforecastnoob]

Thanks for the quick reply!

In a realistic scenario, we of course wouldn't know the future-values of the exogenous, so I get somewhat concerned about data leakage, hence, I don't trust the metrics from the CV as much (if using exog variables in the CV function).
I loose trust because the resulting metric from the CV is when the Exog-forecast are perfect (because it's only using exog- actuals), which isn't realistic I guess.

It would be cool to have this as an option, to use forecasted exog in the CV, instead of actuals. Then you'd have a better idea how accurate the forecasts are expected to be.

Just some thoughts I had, I really like the function otherwise, great job!