A note on forecast reconciliation
本文基于线性投影理论，在最小假设下重新推导了最优协调预测器，阐明了协调误差与初步预测误差的关系，并涵盖了最优线性分解等经典问题。
Forecast reconciliation is a highly effective methodology for improving the predictive accuracy of multiple time series that are subject to linear constraints. This note provides an alternative derivation of the optimal reconciled predictor, based on linear projection arguments, under a set of minimal assumptions on the nature of the reconciliation error. Our result clarifies the relationship between the latter and the preliminary prediction error and encompasses well-known partial reconciliation problems, such as optimal linear disaggregation.