Setting the Forecast Horizon for Competitive Forecast Execution

The Forecast Horizon setting on a competitive forecast group controls how far into the future all methods in that group produce forecasts. It overrides individual method extrapolation settings and ensures that all competing methods forecast the same number of periods. This approach makes the comparison fair and the output consistent.

For example, the competitive forecast group CompClass4-Growing has a Forecast Horizon of 1 Year and includes the methods T MA4 Prev Year, Holt Linear 0.5/0.2, Auto ARIMA, Auto ETS, and Auto THETA. When the group runs on a monthly model, the Forecast Horizon setting overrides the extrapolation settings of the individual methods, causing all methods in the group to generate forecasts for 12 future periods. The method selected as the winner, or the resulting ensemble forecast, then writes 12 forecast periods to the database. Any extrapolation settings defined for individual methods are ignored.

The behavior is to override the ExtraPeriods value defined for individual forecasting methods. For example, if a method is configured to extrapolate 24 periods but the competitive forecast group specifies a Forecast Horizon of 12 periods, the method generates a forecast for only 12 periods. This behavior applies to both the holdout and recalculation phases, ensuring a consistent forecast horizon throughout the competitive selection process. If the Forecast Horizon is set to Per Method Setting, the forecast uses the ExtraPeriods value defined for each individual method.