Resumen
In this paper, we propose a new approach for hierarchical forecasting, which represents the modification of the commonly used ?top down? approach. The proposed method is based on projecting the ratio of bottom and top level series into the future, rather than using average historical proportions and proportions of the historical averages, as in standard top down approaches. Forecasted projections are then used for determining how the ?base? forecasts of the top series will be distributed to the revised forecasts for every series at the bottom level of the hierarchy. Revised forecasts for all series in the hierarchy are obtained in the same manner as in standard top down methodology. In order to estimate the accuracy of the proposed model, the simulation study is performed. Results demonstrate that the ?modified? top down model significantly outperforms standard top down approaches. Beside comparison with the top down approach, the model is compared with the bottom up method, and two newly proposed approaches: top down forecasted proportions and optimal combination approach. Overall, the bottom up method shows the lowest forecasting error. Results also reveal that the modified top down model demonstrates good performance, since its forecasts are close to the forecasts of the top performing model. In the end, we demonstrate our method and others in forecasting the beverage distribution network.