Hey all! A quick question on cross validation. I'm trying to build a (S)ARIMA(X) model for time series having annual seasonality. Im using AutoARIMA but would like to do cross validation for h=28, n_windows=15, step_size=28, refit=False. Meaning I'd like to find best model that predicts most accurately 28 days ahead when looking data from the ~past year. How do I find the best params for p,d,q and P,D,Q using cross validation, or is it even relevant to use cv, and do I just need to put something to AutoARIMA to search best parameters for this particular forecast horizon and seasonality?

Hey. I suggest just using AutoARIMA directly on all of your data, since if you use cross validation you may end up with different parameters in each fold