A0406
Title: Coherent forecast combination for linearly constrained multiple time series
Authors: Daniele Girolimetto - University of Padova (Italy) [presenting]
Tommaso Di Fonzo - University of Padova (Italy)
Abstract: When different, incoherent forecasts of a linearly constrained (i.e. hierarchical) multiple time series are available, both forecast combination and forecast reconciliation may be used to improve the forecast accuracy and achieve coherence in the final forecasts. A regression-based optimal solution is presented to the coherent forecast combination problem for $p > 1$ base (i.e., incoherent) forecast vectors of the same target forecast for a linearly constrained multiple time series. Then, practical issues related to the estimation of the covariance matrix on which the optimal solution is based are discussed. Finally, the effectiveness of the proposed approaches is assessed through two forecasting experiments, using the Australian Energy Market Operator (AEMO) and the Australian Tourism Demand datasets.
