Hello,
in a forecast project I am using tslm in combination with a box cox transformation and the guerrero function in order to get the optimized lamda value.

In order to compare the forecast results of several models (some are transformed and some are not) I need the RMSE, but some of the box cox-transformed models have a RMSE value 0.
The F test seems to be significant, the AICc exists and some of the models have a very good adjusted RSS, so these models seem to be relevant.

In the web I found the hint, that the RMSE of the box cox transformed models sometimes is 0, when the RMSE is calculated before the backtransformation of the forecast. But if I understood the documentation of fable correctly, the forecast values are backtransformed automatically.

Does that mean, that the RMSE (and perhaps AICc etc., too) of a forecast in fable is calculated generally before the backtransformation of the values?
Is that true for simple log tranformations and for other models - like ARIMA - , too?

In this case I have to use the inv_boxcox function after forecasting and use the accuracy function within the backtransformed forecast?

Thanks for a short help.