Data Analysis › Time & Forecasting

Trend

The long-run direction underneath the noise.

Also known as: trend line, upward trend, trend direction

A trend is the slow, long-run direction of a time series: up, down or flat, underneath the weekly wiggles and one-off spikes. Revenue growing 2% a month has an upward trend even in a week it dipped. Naming the trend separately from the noise is the first step of reading any series honestly.

sales:  __/--_/--__/---___/----__/__   (noisy week to week, climbing overall)
trend:  /                              (up and to the right)

A moving average is the quickest way to see it: average each point with its neighbours and the short-term noise mostly cancels, leaving the direction visible. Fitting a straight line through the series does the same job more formally.

The classic mistakes:

• Calling a spike a trend. Three good weeks after a launch is excitement, not a direction. A trend needs enough history to survive the noise — compare against the same period last year, not last week.
• Confusing the slope with the level. “Growing fast” and “big” are different claims. A small new product can grow 50% a month and still be tiny; a mature one growing 1% adds far more revenue.
• Extending the line forever. A trend describes the past. Assuming it continues is forecasting, with all of forecasting’s uncertainty (see forecasting).

Trend is the first of the three time components: direction, repeating seasonality, and leftover noise. Split a series into those three before deciding what moved and why.