Rolling horizon modeling

Rolling horizon modeling enables you to model future forecasts, providing time-phased safety stock and inventory policies. Rolling horizon modeling makes the safety stock placement decision in the "historic horizon", and lets you set time-phased safety stock level in the "planning horizon". For example, you run Safety Stock Optimization on the data in your 12 month historic horizon to determine stocking locations. You then have an additional 3 months of future forecast data. If you were to run Safety Stock Optimization, the new data and the stocking decisions it provides may contradict the original run. With rolling horizon modeling, you can update the model with the new demand and forecast data rolling to the planning horizon without running Safety Stock Optimization again.

When working with rolling horizon modeling, you are dealing with the historic horizon and the planning horizon as shown below:

The historic horizon defines the historical forecast and demand that Safety Stock Optimization uses to determine the inventory placement. It spans from the start of the model horizon to the date selected as Rolling Horizon Start Date. The rolling horizon start date must be within the model horizon. If you populate both the demand and forecast data in the historic horizon, forecast error is calculated. The placement decisions in the historic horizon are long-term and strategic, where the history is long enough to justify them. The placement decision from the historic horizon impacts the planning horizon.

As new forecast data is available, you can shift the rolling horizon start date to reflect the change in the new planning horizon, then run Rolling Horizon Modeling to roll the safety stock decisions forward: