A Crash Course in Forecast Reconciliation
This is a two-hour guest lecture delivered in the Forecasting and Decision Making course at the master’s level at Hunan University. Students are assumed to have a background in statistics and to be familiar with univariate forecasting methods, but not with hierarchical forecasting or forecast reconciliation.
The lecture introduces the fundamentals of hierarchical forecasting, including traditional approaches and cross-sectional forecast reconciliation. It also presents key theoretical results, such as the unbiasedness property and forecast error variance, alongside geometric interpretations that provide additional intuition. The session concludes with a brief overview of extensions, including temporal and cross-temporal reconciliation, probabilistic forecast reconciliation, and Forecast Linear Augmented Projection (FLAP).