quant5
commented
2 months ago There won't be a one-size fits all answer to this question. Some models e.g., ARIMA require stationary and attempt to handle the differencing for you. Depending on your data, you have to additional differencing, like seasonal differencing, to achieve stationarity. Other models like Holt-Winters do not require the data to be stationary at all.
A rule of thumb is that if the model is designed to accept data with non-stationary components like trend and seasonality, you don't need to handle them beforehand.
Suggested reading: https://otexts.com/fpp3/stationarity.html
Description
This is not quite clear from the library documentation; does non-stationarity require explicit handling via e.g. differencing? If the answer depends on the model, please let me know which model requires such an approach. Many thanks!
Link
No response