If with AutoARIMA I want to constrain the predictions to be positive (>= 0), are there better alternatives than log-transforming the data (after summing epsilon=1 to handle values equal to zero) before launching AutoARIMA? Thanks!
d
Dihong Huang
06/02/2023, 9:56 PM
What about manually mask the results?
m
Manuel
06/03/2023, 10:58 AM
The problem is that sometimes AutoARIMA (especially with high seasonal orders) captures negative trends that do not exist, and the forecasts become all negative