Demand Forecasting Model (forecast)
Create forecasts with automatic selection or manual model choices across exponential smoothing, moving average, and linear regression.
Quick Start
Which method should I choose?
| Your data looks like | Try |
|---|---|
| Flat, no clear direction | Simple Exponential Smoothing or Moving Average |
| Steady rise or fall | Holt Linear Trend or Linear Regression |
| A repeating pattern (months, quarters) | Seasonal or Holt-Winters |
| Seasonal swings that grow with the level | Multiplicative versions (all values must be above zero) |
| Not sure | Automatic, which compares the eligible models for you |
How much history do I need?
- At least 8 rows for any model.
- Seasonal models need two full cycles, for example 24 rows of monthly data with a seasonal period of 12.
- Moving Average needs at least the window size plus 2 rows.
Reading the accuracy metrics
- MAD (MAE): the average size of the error, in the same units as your data.
- MSE and RMSE: penalise large misses more heavily. RMSE is back in your data units.
- MAPE %: the average error as a percentage of the actual value. Rows where the actual value is zero or negative are skipped.
- Lower is better, but only compare models on the same metric and the same validation mode.
Here is the actual value, is the forecast and is the number of scored periods.
- With Holdout validation the metrics are out-of-sample, so they are a more honest guide than in-sample fit.
Tips
- With Holdout validation, the Decomposition table shows the Holdout forecast and Holdout error for each held-out period.
- Typical seasonal periods: 12 for monthly data, 4 for quarterly data, 7 for daily data with a weekly pattern.
- Moving Average forecasts flatten out over longer horizons. If you expect growth to continue, use a trend model.
- Confidence bounds are estimated from past residuals and widen with the forecast horizon. They are not guarantees.
- Use the forecast as the demand input to the related inventory, production and workforce planning calculators.
Click here to Learn more about: Demand Forecasting Model