Trend Component

The trend captures where a series is heading once short-term movements are set aside.

Most business questions about growth, decline and stability are questions about the trend.

A trend can be linear, curved, or broken into different stretches by events such as a launch.

A simple way to see it is to average each full year, which removes the ups and downs within it.

Fitting a straight line over time gives a rough trend and a single growth figure per period.

Growth that compounds, such as 5% a month, looks curved on a normal axis and straight on a log scale.

Decomposition methods split a series into trend, seasonal and leftover parts at once.

The leftover part should look like random noise; any pattern left in it means something was missed.

Trends found in short series are fragile, since a few unusual months can create or hide one.

Changes in how data is collected, such as a new tracking tool, can create a fake trend.

Comparing a trend with a related series, such as orders against site visits, helps judge whether it is real.

Extending a trend assumes the forces behind it will continue, which is a business judgement, not a statistical one.

Saturation is a common reason trends bend, as when a product reaches most of its possible customers.

In practice, forecasts often blend a trend estimate with judgement about upcoming changes.

The usual mistake is reading a seasonal rise, such as December sales, as a change in the trend.