Dirichlet Proportions Model for Hierarchically Coherent Probabilistic Forecasting
This addresses the need for coherent probabilistic predictions in practical forecasting applications, such as retail or finance, and is incremental as it builds on classical top-down strategies with deep learning enhancements.
The paper tackles the problem of probabilistic, hierarchically coherent forecasting for time series arranged in a tree hierarchy by proposing an end-to-end deep probabilistic model based on a top-down strategy, resulting in significant improvements of up to 26% compared to state-of-the-art baselines on public datasets.
Probabilistic, hierarchically coherent forecasting is a key problem in many practical forecasting applications -- the goal is to obtain coherent probabilistic predictions for a large number of time series arranged in a pre-specified tree hierarchy. In this paper, we present an end-to-end deep probabilistic model for hierarchical forecasting that is motivated by a classical top-down strategy. It jointly learns the distribution of the root time series, and the (dirichlet) proportions according to which each parent time-series is split among its children at any point in time. The resulting forecasts are naturally coherent, and provide probabilistic predictions over all time series in the hierarchy. We experiment on several public datasets and demonstrate significant improvements of up to 26% on most datasets compared to state-of-the-art baselines. Finally, we also provide theoretical justification for the superiority of our top-down approach compared to the more traditional bottom-up modeling.