This message was deleted.

Hey. What do you mean by "been fitted well"?

Sometimes, the training data is too noisy or lacks a clear pattern, so the model can not describe such data well. The question is how to detect such cases?

You can look at the training errors. Which method are you using?

the method is AutoARIMA

I mean fit, forecast, cross_validation

forecast

You can use:

sf = StatsForecast(...)
sf.forecast(..., fitted=True) # this stores the in sample predictions
train_preds = sf.forecast_fitted_values()

You can then use the

y

and

AutoARIMA

columns of train_preds to compute the training error