Before you configure your Forecaster forecasts, decide which time dimension type to apply. You can use either the built-in time dimension, or a custom one.
You also need to consider:
- The types of data (examples: financial results, HR information, or inventory levels).
- The frequency of the forecast to run.
The model calendar type you select depends on whether your source data is monthly, weekly, or daily. For more information about each calendar type, see Set the model calendar.
Note: Forecaster doesn't support calendar periods in the data, such as P1 FY22.
Monthly
If your source data is monthly, use the Calendar Months/Quarters/Years calendar type. If you use custom months, you must follow the calendar/month convention.
Weekly
If your source data is weekly, use one of these calendar types:
- Weeks: 4-4-4, 4-5-4 or 5-4-4
- Weeks: General
If your data uses a custom fiscal year, use the Weeks: General calendar type. - Weeks: 13 4-week Periods
Daily
If your source data is daily, you can choose any of the Weeks calendar types.
For each line item in the historical and related data modules, make sure to set the Time Scale to Day. You must also ensure the forecast results module, and the optional Insights module, are also set to Day.
Custom time dimension
Forecaster also supports an optional custom ("fake") time dimension. Example: you might use custom time settings to account for longer or shorter historical data spans. The same frequencies, daily, weekly, or monthly apply.
- The time dimension must be in the form of a list of timestamps, YYYY-MM-DD (supported separators: dash (-), forward slash (/), period (.), colon (:)).
- When you use a monthly custom time dimension, the day (DD) in YYYY-MM-DD must be the first day of the month (01).
- Forecaster supports list subsets within custom time dimensions. If you use a custom list, be sure to include all future dates to cover the full length of the forecast horizon.
