Choose a smoother without erasing the event

Why this lab Choose a smoother that keeps the real event visible.

Wider windows suppress short-term variation, flatten peaks and lose more periods at both ends. These averages are centered, so widening them does not shift the line sideways — a trailing average would, and that is the lag cost centering avoids. Compare three windows on the same series, then select the smallest window that clarifies trend without hiding the intervention shock.

A smoother is a transformation, not the data. A moving average is a summary you chose, with a window you chose; it is not the true series hiding underneath. A one-period spike in the raw values is a recorded observation. Keep it, annotate it and investigate it; a smoother may shrink how large it looks, but deleting the point because the smoothed line does not show it would misreport the data.