Forecasting is the process of making statements about events whose actual outcomes have not yet been observed. It is used for decades in different fields like climate, crime, health, business... Although the purpose of different forecasting systems is not the same, in general, they help decision-makers to make appropriate plans for future likely events. As the nature of forecasting methods and measures are often quantitative, these predictive analytics systems usually use a data warehouse to store data and OLAP tools to visualize query/simulation results. A specific feature of forecasting systems regarding predictions analysis is backward propagation of updates, which is the computation of the impact, on raw data, of modifications performed on summaries. In data warehouses, some methods propagate updates over hierarchies when modifications are performed on data sources. However, so far, very few works have been devoted to update propagation from summaries to raw data. This paper proposes an algorithm called PAM (Propagation of Aggregate-based Modification), to efficiently propagate modifications performed on summaries to raw data, and then to other summaries. Ex- periments have been conducted on an operational application.