Abstract
When organizations plan to enter a new market or to expand their business to new locations, they need sales time-series or final consumers scanner data in time and geographical dis-aggregation to implement their desired marketing strategy. However, such data are hardly available, which may seriously undermine the organizations' dis-aggregated forecasting efforts. Based on agglomeration and gravitational theories, we propose a new forecast reconciliation approach that distributes and reconciles forecasts to lower levels of aggregation, when no actual dis-aggregated sales data or historical sales proportions are available. We combine the deep learning technique Long Short Term Memory (LSTM) applied to time-series forecasting with a new gravitational model approach and validate our method with real sales data of two different companies from two different countries. The results show that our proposed reconciliation approach based on easily or freely available data has a similar or better performance than the benchmark approaches using proprietary sales data at all levels of the companies' channel hierarchy.