Creating forecasts

You can create a forecast using the Forecasts editor. You use the Group and Model pages of the Forecasts editor to set up a forecasting model.

About this task

To create a forecast, perform the following procedure:


  1. Open the Forecasts editor in one of the following ways:
    • With an active query in the Editor window, select Query > Transfer To > Forecast
    • Select File > New > Other

      The New Wizard wizard opens. From the QMF Objects folder select the Forecasts wizard

    • Click the New Forecast toolbar icon.
    The Forecast editor opens in a separate tab.
  2. Select the query or table that will be used to source the historical data.
  3. Specify the Date Parameters options.
  4. Specify the forecasting model's grouping.

    The Grouping Hierarchy is used to specify what values are grouped and how they are ordered. For example, in wine sales, group by wine type and then location or by location then wine type.

  5. Specify the query column that contains the values that will be forecast and the method of aggregation.
  6. Specify the forecasting model's construction strategy, and distribution strategy:
    • The Construction Strategy is used to specify either a top-down or bottom-up approach, where the root node is at the top and the leaf node is at the bottom of a hierarchical tree diagram.
    • The Distribution Strategy specifies how the forecast values are distributed using the top-down construction strategy (e.g., from root to leaf nodes).
  7. Specify Forecasting Models options including the forecasting models that are used, and their associated parameters.
  8. Click Run Forecast on the toolbar.
    The graphed forecast is displayed in the Results editor.
  9. Experiment with one or more forecasting models to determine the best fit. Validation methods include:
    • Validating the forecasting model using performance measures.
      There are five performance measures that may be used as comparative criteria, including:
      • Cumulative Forecast Error
      • Mean Absolute Deviation
      • Mean Square Error
      • Mean Absolute Percent Error
      • Tracking Signal
      In most cases the Mean Square Error is used as the comparative criteria.
    • Validating the forecasting model time series elements in accordance with any observed trend, seasonality, and cyclicity.

      In the case of trend, an observed trend at a lower node may influence the choice of forecasting model for the entire forecast. For example, when marketing a new product it may be wiser to base the forecasting model on a smaller, more representative demographic with observable trends than use a larger, more diverse demographic with distorted or no observable trends.

  10. Save the forecast to a file or to the repository.
    Note: You can use saved forecasts as query objects when creating analytical queries, prompt hierarchies, drill down paths, quick reports, and visual projects.