This message was deleted.

Hi @Afiq Johari! Thanks for sharing your approach. It is completely valid, retraining the best model after hyperparameter optimization with new data is a very common practice, specially given that your test period is so long. The only issue I see is that your validation period is smaller than the horizon. You will need to increase it to at least match the horizon.

@Cristian (Nixtla) thanks for your feedback! The test period (5 years or 20 quarters) reflect our actual forecast horizon. I'll try to match the validation and test horizon. Will share if I found anything interesting.