Glosarium
Contents

Data Analysis › Time & Forecasting

Moving Average

Smoothing a series by averaging each point with its neighbours.

Data Analyst · Mid-level

Also known as: moving average, rolling average, 7-day average

A moving average replaces each point in a time series with the average of the points around it — the last 7 days, say — and the jagged line turns into a readable one. Day-to-day noise cancels out while slower movement survives, so the trend and turning points become visible. The 7-day average on every COVID-era dashboard was a moving average doing exactly this job.

daily:   5  9  4  12  6  15  3  ...
7-day avg:       7.7  8.4  ...        (each point = mean of surrounding days)

Window size is the whole trade-off. A short window follows turns quickly but leaves noise; a long window is smooth but lags — a 30-day average confirms a downturn weeks after it started. There is no correct size, only the one matched to the question: detecting a change wants short, showing direction wants long.

The classic mistakes:

  • Averaging away the signal. A 30-day window on data with strong weekly seasonality hides both the season and fast breaks. Check the raw series alongside the smoothed one.
  • Treating the smoothed line as data. The average lags reality by design; the most recent smoothed value describes the middle of its window, not today. Never read “current value” off a moving average.
  • Centered vs trailing confusion. A centered average uses future points and cannot be computed for the latest days; a trailing average can, but lags more. Dashboards should use trailing and say so.
  • Smoothing before alerting. Alerts on smoothed values fire late. Detect on raw data, display smoothed.

When not to use it: when each point matters (alerts, exact values) or when the series needs modelling rather than viewing — then reach for exponential smoothing or a real forecast. Moving averages are for eyes, not for decisions made by machines.