Demand forecasting
Forecast Backtesting
By the Supply Chain Manager team · Updated October 16, 2026
Backtesting checks a forecasting method by pretending to forecast the past: train on older history, predict a later period and compare with what really sold.
Rolling-origin backtesting repeats this several times, moving the cut-off forward each time, so the result does not hinge on one lucky or unlucky period. It is the fair way to choose between methods for a SKU before trusting one with real purchase orders.
Backtesting example
Illustrative: with a year of daily history, train on the first nine months and forecast the next 14 days, then move the cut-off 14 days later and repeat. Averaging the error over several such windows shows which method holds up, not just which one got lucky once.
Why backtesting matters for distributors
Different SKUs suit different methods. Backtesting each one lets the data, not habit, pick the method, and shows honestly how much accuracy to expect.
Common backtesting mistakes
- Testing a method on the same data it was fitted to.
- Picking the most complex model when a simpler one scores about the same.
Backtesting in Supply Chain Manager
Supply Chain Manager runs rolling-origin backtests over 14-day windows for each SKU, compares methods by WAPE and keeps the simpler model on close calls.
Open the live demoExamples on this page use made-up round numbers to show the calculation. They are not benchmarks or results from real businesses.