Why Stationarity Alone Does Not Make a Series Forecastable

Summary

The document raises a conceptual question about the relationship between stationarity and forecasting. It observes that white noise is stationary yet is generally treated as unpredictable, while time-series forecasting often begins by transforming data to achieve stationarity. The author asks whether stationarity therefore fails to guarantee forecastability, and why it is useful in forecasting workflows at all.

No answer, model, examples, or empirical evidence are included, so the issue remains open in the document. It is best read as a prompt to distinguish stable statistical properties over time from predictable dependence in observations. Making a series stationary may support the assumptions of a forecasting method, but this document does not explain those assumptions or describe a forecasting procedure.

Key ideas

• The document contrasts white noise stationarity with its lack of useful predictability.
• It asks whether stationarity is sufficient for forecastability.
• It questions why stationarity is commonly sought in time-series forecasting.
• No explanation or evidence answering these questions is provided.

Tags

# white noise not forecastable ? stationarity doesn't imply forecastability?


# white noise not forecastable ? stationarity doesn't imply forecastability?












We know that white noise isn't forecastable because of its random aspect. White noise is also stationary, and which is confusing me, is that we always try to make a serie stationary to make forecasts, hence I don't understant why white noise stationarity doesn't make it forecastable...

The only conclusion I see is that stationarity, doesn't imply forecastability. Thus, why are we always trying to make series stationaire, if it doesn't imply forecastability ?

Could someone please help me understand these notions ?

Thank you very much Habib

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.