MerQuR+Statistics Glossary

Statistics Glossary → Moving-average and exponential moving-average smoothing

Moving-average and exponential moving-average smoothingconcept

Smoothing methods that suppress short-term fluctuations in a time series by averaging nearby observations at each point, making the trend visible.

The simple moving average MA(w) is the average of the last w observations (trailing) or of the observations on both sides of the point (centred). A centred moving average introduces no lag; if w is even (e.g. 12 for monthly data), a two-stage 2 × w average is used so that the average lines up with the observation times. If the window equals the seasonal length, seasonal fluctuation also disappears. A trailing average uses only the past and therefore lags behind the trend.

The exponential moving average (EMA) uses all past observations with weights that decline exponentially going back in time: s_t = α · x_t + (1 - α) · s_{t-1}. The larger α, the more weight recent observations get and the less the series is smoothed. These smoothing averages should not be confused with the 'moving average (MA) term' of an ARIMA model, which describes the effect of past random shocks. A smoothed series is less reliable at its end points and shows turning points with a delay.

An everyday example

A pharmacy's daily sales swing a lot between weekdays and weekends. Taking a 7-day moving average removes the weekly wave and clearly shows sales slowly rising during the winter months. The EMA, which gives more weight to recent days, reflects this rise a little earlier.

In MerQur

The 'Smoothing (MA window / EMA α), centred' option of the Time Series analysis smooths the series with a trailing MA(w), a centred MA (2 × w for even w) and an EMA.

Analyses where this concept appears

Time Series

Related concepts

Smoothing parameter (λ, span)