Appearance
Forecasting
Also known as
- forecast settings
- forecasting methods
- default method
- smoothing alpha
- moving average window
- history months
- horizon
- backtest holdout
- monthly recompute
- accuracy window
Forecasting settings decide which methods every forecast run compares, how much history it learns from, how far ahead it forecasts and whether it recomputes by itself every month. A demand planning lead or administrator sets them once and revisits them when accuracy drops.
Change the forecasting settings
You need Permission: supply_chain.settings.manage to change them; with
supply_chain.settings.view you see them read-only.- Open Supply chainSettingsForecasting.
- Change the fields below.
- Select Save settings. The message Settings saved appears.
New values apply from the next forecast run; existing forecasts do not change until then.
| Field | What it does | Default |
|---|---|---|
| Enabled methods | Moving average, Exponential smoothing, Holt-Winters (seasonal). Every run fits and backtests the enabled ones. | All three |
| Default method | Used when automatic selection is off. | Holt-Winters (seasonal) |
| Smoothing alpha (0 = fitted) | Weight of the latest month in exponential smoothing; 0 lets each run find the best value. | 0 |
| Moving average window (months) | Months averaged by the moving average. | 3 |
| Pick the method with the lowest backtest MAPE per series | On: each item × warehouse gets its best method. Off: the default method everywhere (unless a planner fixed one). | On |
| History (months) | Complete months of demand history a run learns from. | 24 |
| Horizon (months) | Months forecast from the current month. | 12 |
| Backtest holdout (months) | Last months re-forecast one step ahead to measure MAPE. | 6 |
| Accuracy window (months) | Months measured on Forecast accuracy and in the KPI. | 6 |
| Months shared with vendors | Months of forecast a vendor sees in the vendor portal. | 6 |
| Recompute the forecast on the first of every month | Runs the forecast for every series on the 1st (trigger Monthly job). | On |
Holt-Winters needs two full seasons (24 months) of history; with less it falls back to a trend-only model.
