Validate forecasts with rolling origins
Use rolling-origin validation to evaluate a forecast without letting future time periods inform the past.
A weekly SKU forecast needs a four-week-ahead error estimate for replenishment planning. Rolling-origin validation: train on the past, test on the future, repeat across origins Randomly splitting a time series leaks future seasonality and promotion effects into training. Define horizon Forecast four weeks ahead because purchase orders must be placed four weeks before demand. The validation horizon must match the decision horizon. Choose origins Use origin dates at the end of March, April, May, and June. Multiple origins test different market conditions instead of one lucky split. Train only on prior data For the April origin, train through April only…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in