11.2 Preparing time-series data

Before fitting a forecasting method, determine:

• what one observation represents;
• whether periods are equally spaced;
• whether the series is a count, rate, average, or total;
• whether partial periods are present;
• whether definitions changed;
• whether missing periods mean zero or unrecorded data;
• whether unusual events should be modelled, adjusted, or retained; and
• how far into the future the forecast must extend.

Visualize the series before modelling. Look for trend, seasonality, level changes, changing variation, outliers, and structural breaks. A method that performed well under one operating system may not remain valid after a major policy, facility, or data-collection change.